Commit | Line | Data |
---|---|---|
ddbcc7e8 PM |
1 | /* |
2 | * kernel/cgroup.c | |
3 | * | |
4 | * Generic process-grouping system. | |
5 | * | |
6 | * Based originally on the cpuset system, extracted by Paul Menage | |
7 | * Copyright (C) 2006 Google, Inc | |
8 | * | |
9 | * Copyright notices from the original cpuset code: | |
10 | * -------------------------------------------------- | |
11 | * Copyright (C) 2003 BULL SA. | |
12 | * Copyright (C) 2004-2006 Silicon Graphics, Inc. | |
13 | * | |
14 | * Portions derived from Patrick Mochel's sysfs code. | |
15 | * sysfs is Copyright (c) 2001-3 Patrick Mochel | |
16 | * | |
17 | * 2003-10-10 Written by Simon Derr. | |
18 | * 2003-10-22 Updates by Stephen Hemminger. | |
19 | * 2004 May-July Rework by Paul Jackson. | |
20 | * --------------------------------------------------- | |
21 | * | |
22 | * This file is subject to the terms and conditions of the GNU General Public | |
23 | * License. See the file COPYING in the main directory of the Linux | |
24 | * distribution for more details. | |
25 | */ | |
26 | ||
27 | #include <linux/cgroup.h> | |
28 | #include <linux/errno.h> | |
29 | #include <linux/fs.h> | |
30 | #include <linux/kernel.h> | |
31 | #include <linux/list.h> | |
32 | #include <linux/mm.h> | |
33 | #include <linux/mutex.h> | |
34 | #include <linux/mount.h> | |
35 | #include <linux/pagemap.h> | |
a424316c | 36 | #include <linux/proc_fs.h> |
ddbcc7e8 PM |
37 | #include <linux/rcupdate.h> |
38 | #include <linux/sched.h> | |
817929ec | 39 | #include <linux/backing-dev.h> |
ddbcc7e8 PM |
40 | #include <linux/seq_file.h> |
41 | #include <linux/slab.h> | |
42 | #include <linux/magic.h> | |
43 | #include <linux/spinlock.h> | |
44 | #include <linux/string.h> | |
bbcb81d0 | 45 | #include <linux/sort.h> |
81a6a5cd | 46 | #include <linux/kmod.h> |
ddbcc7e8 PM |
47 | #include <asm/atomic.h> |
48 | ||
81a6a5cd PM |
49 | static DEFINE_MUTEX(cgroup_mutex); |
50 | ||
ddbcc7e8 PM |
51 | /* Generate an array of cgroup subsystem pointers */ |
52 | #define SUBSYS(_x) &_x ## _subsys, | |
53 | ||
54 | static struct cgroup_subsys *subsys[] = { | |
55 | #include <linux/cgroup_subsys.h> | |
56 | }; | |
57 | ||
58 | /* | |
59 | * A cgroupfs_root represents the root of a cgroup hierarchy, | |
60 | * and may be associated with a superblock to form an active | |
61 | * hierarchy | |
62 | */ | |
63 | struct cgroupfs_root { | |
64 | struct super_block *sb; | |
65 | ||
66 | /* | |
67 | * The bitmask of subsystems intended to be attached to this | |
68 | * hierarchy | |
69 | */ | |
70 | unsigned long subsys_bits; | |
71 | ||
72 | /* The bitmask of subsystems currently attached to this hierarchy */ | |
73 | unsigned long actual_subsys_bits; | |
74 | ||
75 | /* A list running through the attached subsystems */ | |
76 | struct list_head subsys_list; | |
77 | ||
78 | /* The root cgroup for this hierarchy */ | |
79 | struct cgroup top_cgroup; | |
80 | ||
81 | /* Tracks how many cgroups are currently defined in hierarchy.*/ | |
82 | int number_of_cgroups; | |
83 | ||
84 | /* A list running through the mounted hierarchies */ | |
85 | struct list_head root_list; | |
86 | ||
87 | /* Hierarchy-specific flags */ | |
88 | unsigned long flags; | |
81a6a5cd PM |
89 | |
90 | /* The path to use for release notifications. No locking | |
91 | * between setting and use - so if userspace updates this | |
92 | * while child cgroups exist, you could miss a | |
93 | * notification. We ensure that it's always a valid | |
94 | * NUL-terminated string */ | |
95 | char release_agent_path[PATH_MAX]; | |
ddbcc7e8 PM |
96 | }; |
97 | ||
98 | ||
99 | /* | |
100 | * The "rootnode" hierarchy is the "dummy hierarchy", reserved for the | |
101 | * subsystems that are otherwise unattached - it never has more than a | |
102 | * single cgroup, and all tasks are part of that cgroup. | |
103 | */ | |
104 | static struct cgroupfs_root rootnode; | |
105 | ||
106 | /* The list of hierarchy roots */ | |
107 | ||
108 | static LIST_HEAD(roots); | |
817929ec | 109 | static int root_count; |
ddbcc7e8 PM |
110 | |
111 | /* dummytop is a shorthand for the dummy hierarchy's top cgroup */ | |
112 | #define dummytop (&rootnode.top_cgroup) | |
113 | ||
114 | /* This flag indicates whether tasks in the fork and exit paths should | |
115 | * take callback_mutex and check for fork/exit handlers to call. This | |
116 | * avoids us having to do extra work in the fork/exit path if none of the | |
117 | * subsystems need to be called. | |
118 | */ | |
119 | static int need_forkexit_callback; | |
120 | ||
121 | /* bits in struct cgroup flags field */ | |
122 | enum { | |
81a6a5cd | 123 | /* Control Group is dead */ |
ddbcc7e8 | 124 | CONT_REMOVED, |
81a6a5cd PM |
125 | /* Control Group has previously had a child cgroup or a task, |
126 | * but no longer (only if CONT_NOTIFY_ON_RELEASE is set) */ | |
127 | CONT_RELEASABLE, | |
128 | /* Control Group requires release notifications to userspace */ | |
129 | CONT_NOTIFY_ON_RELEASE, | |
ddbcc7e8 PM |
130 | }; |
131 | ||
132 | /* convenient tests for these bits */ | |
133 | inline int cgroup_is_removed(const struct cgroup *cont) | |
134 | { | |
135 | return test_bit(CONT_REMOVED, &cont->flags); | |
136 | } | |
137 | ||
138 | /* bits in struct cgroupfs_root flags field */ | |
139 | enum { | |
140 | ROOT_NOPREFIX, /* mounted subsystems have no named prefix */ | |
141 | }; | |
142 | ||
81a6a5cd PM |
143 | inline int cgroup_is_releasable(const struct cgroup *cont) |
144 | { | |
145 | const int bits = | |
146 | (1 << CONT_RELEASABLE) | | |
147 | (1 << CONT_NOTIFY_ON_RELEASE); | |
148 | return (cont->flags & bits) == bits; | |
149 | } | |
150 | ||
151 | inline int notify_on_release(const struct cgroup *cont) | |
152 | { | |
153 | return test_bit(CONT_NOTIFY_ON_RELEASE, &cont->flags); | |
154 | } | |
155 | ||
ddbcc7e8 PM |
156 | /* |
157 | * for_each_subsys() allows you to iterate on each subsystem attached to | |
158 | * an active hierarchy | |
159 | */ | |
160 | #define for_each_subsys(_root, _ss) \ | |
161 | list_for_each_entry(_ss, &_root->subsys_list, sibling) | |
162 | ||
163 | /* for_each_root() allows you to iterate across the active hierarchies */ | |
164 | #define for_each_root(_root) \ | |
165 | list_for_each_entry(_root, &roots, root_list) | |
166 | ||
81a6a5cd PM |
167 | /* the list of cgroups eligible for automatic release. Protected by |
168 | * release_list_lock */ | |
169 | static LIST_HEAD(release_list); | |
170 | static DEFINE_SPINLOCK(release_list_lock); | |
171 | static void cgroup_release_agent(struct work_struct *work); | |
172 | static DECLARE_WORK(release_agent_work, cgroup_release_agent); | |
173 | static void check_for_release(struct cgroup *cont); | |
174 | ||
817929ec PM |
175 | /* Link structure for associating css_set objects with cgroups */ |
176 | struct cg_cgroup_link { | |
177 | /* | |
178 | * List running through cg_cgroup_links associated with a | |
179 | * cgroup, anchored on cgroup->css_sets | |
180 | */ | |
181 | struct list_head cont_link_list; | |
182 | /* | |
183 | * List running through cg_cgroup_links pointing at a | |
184 | * single css_set object, anchored on css_set->cg_links | |
185 | */ | |
186 | struct list_head cg_link_list; | |
187 | struct css_set *cg; | |
188 | }; | |
189 | ||
190 | /* The default css_set - used by init and its children prior to any | |
191 | * hierarchies being mounted. It contains a pointer to the root state | |
192 | * for each subsystem. Also used to anchor the list of css_sets. Not | |
193 | * reference-counted, to improve performance when child cgroups | |
194 | * haven't been created. | |
195 | */ | |
196 | ||
197 | static struct css_set init_css_set; | |
198 | static struct cg_cgroup_link init_css_set_link; | |
199 | ||
200 | /* css_set_lock protects the list of css_set objects, and the | |
201 | * chain of tasks off each css_set. Nests outside task->alloc_lock | |
202 | * due to cgroup_iter_start() */ | |
203 | static DEFINE_RWLOCK(css_set_lock); | |
204 | static int css_set_count; | |
205 | ||
206 | /* We don't maintain the lists running through each css_set to its | |
207 | * task until after the first call to cgroup_iter_start(). This | |
208 | * reduces the fork()/exit() overhead for people who have cgroups | |
209 | * compiled into their kernel but not actually in use */ | |
210 | static int use_task_css_set_links; | |
211 | ||
212 | /* When we create or destroy a css_set, the operation simply | |
213 | * takes/releases a reference count on all the cgroups referenced | |
214 | * by subsystems in this css_set. This can end up multiple-counting | |
215 | * some cgroups, but that's OK - the ref-count is just a | |
216 | * busy/not-busy indicator; ensuring that we only count each cgroup | |
217 | * once would require taking a global lock to ensure that no | |
b4f48b63 PM |
218 | * subsystems moved between hierarchies while we were doing so. |
219 | * | |
220 | * Possible TODO: decide at boot time based on the number of | |
221 | * registered subsystems and the number of CPUs or NUMA nodes whether | |
222 | * it's better for performance to ref-count every subsystem, or to | |
223 | * take a global lock and only add one ref count to each hierarchy. | |
224 | */ | |
817929ec PM |
225 | |
226 | /* | |
227 | * unlink a css_set from the list and free it | |
228 | */ | |
81a6a5cd | 229 | static void unlink_css_set(struct css_set *cg) |
b4f48b63 | 230 | { |
817929ec PM |
231 | write_lock(&css_set_lock); |
232 | list_del(&cg->list); | |
233 | css_set_count--; | |
234 | while (!list_empty(&cg->cg_links)) { | |
235 | struct cg_cgroup_link *link; | |
236 | link = list_entry(cg->cg_links.next, | |
237 | struct cg_cgroup_link, cg_link_list); | |
238 | list_del(&link->cg_link_list); | |
239 | list_del(&link->cont_link_list); | |
240 | kfree(link); | |
241 | } | |
242 | write_unlock(&css_set_lock); | |
81a6a5cd PM |
243 | } |
244 | ||
245 | static void __release_css_set(struct kref *k, int taskexit) | |
246 | { | |
247 | int i; | |
248 | struct css_set *cg = container_of(k, struct css_set, ref); | |
249 | ||
250 | unlink_css_set(cg); | |
251 | ||
252 | rcu_read_lock(); | |
253 | for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) { | |
254 | struct cgroup *cont = cg->subsys[i]->cgroup; | |
255 | if (atomic_dec_and_test(&cont->count) && | |
256 | notify_on_release(cont)) { | |
257 | if (taskexit) | |
258 | set_bit(CONT_RELEASABLE, &cont->flags); | |
259 | check_for_release(cont); | |
260 | } | |
261 | } | |
262 | rcu_read_unlock(); | |
817929ec | 263 | kfree(cg); |
b4f48b63 PM |
264 | } |
265 | ||
81a6a5cd PM |
266 | static void release_css_set(struct kref *k) |
267 | { | |
268 | __release_css_set(k, 0); | |
269 | } | |
270 | ||
271 | static void release_css_set_taskexit(struct kref *k) | |
272 | { | |
273 | __release_css_set(k, 1); | |
274 | } | |
275 | ||
817929ec PM |
276 | /* |
277 | * refcounted get/put for css_set objects | |
278 | */ | |
279 | static inline void get_css_set(struct css_set *cg) | |
280 | { | |
281 | kref_get(&cg->ref); | |
282 | } | |
283 | ||
284 | static inline void put_css_set(struct css_set *cg) | |
285 | { | |
286 | kref_put(&cg->ref, release_css_set); | |
287 | } | |
288 | ||
81a6a5cd PM |
289 | static inline void put_css_set_taskexit(struct css_set *cg) |
290 | { | |
291 | kref_put(&cg->ref, release_css_set_taskexit); | |
292 | } | |
293 | ||
817929ec PM |
294 | /* |
295 | * find_existing_css_set() is a helper for | |
296 | * find_css_set(), and checks to see whether an existing | |
297 | * css_set is suitable. This currently walks a linked-list for | |
298 | * simplicity; a later patch will use a hash table for better | |
299 | * performance | |
300 | * | |
301 | * oldcg: the cgroup group that we're using before the cgroup | |
302 | * transition | |
303 | * | |
304 | * cont: the cgroup that we're moving into | |
305 | * | |
306 | * template: location in which to build the desired set of subsystem | |
307 | * state objects for the new cgroup group | |
308 | */ | |
309 | ||
310 | static struct css_set *find_existing_css_set( | |
311 | struct css_set *oldcg, | |
312 | struct cgroup *cont, | |
313 | struct cgroup_subsys_state *template[]) | |
b4f48b63 PM |
314 | { |
315 | int i; | |
817929ec PM |
316 | struct cgroupfs_root *root = cont->root; |
317 | struct list_head *l = &init_css_set.list; | |
318 | ||
319 | /* Built the set of subsystem state objects that we want to | |
320 | * see in the new css_set */ | |
321 | for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) { | |
322 | if (root->subsys_bits & (1ull << i)) { | |
323 | /* Subsystem is in this hierarchy. So we want | |
324 | * the subsystem state from the new | |
325 | * cgroup */ | |
326 | template[i] = cont->subsys[i]; | |
327 | } else { | |
328 | /* Subsystem is not in this hierarchy, so we | |
329 | * don't want to change the subsystem state */ | |
330 | template[i] = oldcg->subsys[i]; | |
331 | } | |
332 | } | |
333 | ||
334 | /* Look through existing cgroup groups to find one to reuse */ | |
335 | do { | |
336 | struct css_set *cg = | |
337 | list_entry(l, struct css_set, list); | |
338 | ||
339 | if (!memcmp(template, cg->subsys, sizeof(cg->subsys))) { | |
340 | /* All subsystems matched */ | |
341 | return cg; | |
342 | } | |
343 | /* Try the next cgroup group */ | |
344 | l = l->next; | |
345 | } while (l != &init_css_set.list); | |
346 | ||
347 | /* No existing cgroup group matched */ | |
348 | return NULL; | |
349 | } | |
350 | ||
351 | /* | |
352 | * allocate_cg_links() allocates "count" cg_cgroup_link structures | |
353 | * and chains them on tmp through their cont_link_list fields. Returns 0 on | |
354 | * success or a negative error | |
355 | */ | |
356 | ||
357 | static int allocate_cg_links(int count, struct list_head *tmp) | |
358 | { | |
359 | struct cg_cgroup_link *link; | |
360 | int i; | |
361 | INIT_LIST_HEAD(tmp); | |
362 | for (i = 0; i < count; i++) { | |
363 | link = kmalloc(sizeof(*link), GFP_KERNEL); | |
364 | if (!link) { | |
365 | while (!list_empty(tmp)) { | |
366 | link = list_entry(tmp->next, | |
367 | struct cg_cgroup_link, | |
368 | cont_link_list); | |
369 | list_del(&link->cont_link_list); | |
370 | kfree(link); | |
371 | } | |
372 | return -ENOMEM; | |
373 | } | |
374 | list_add(&link->cont_link_list, tmp); | |
375 | } | |
376 | return 0; | |
377 | } | |
378 | ||
379 | static void free_cg_links(struct list_head *tmp) | |
380 | { | |
381 | while (!list_empty(tmp)) { | |
382 | struct cg_cgroup_link *link; | |
383 | link = list_entry(tmp->next, | |
384 | struct cg_cgroup_link, | |
385 | cont_link_list); | |
386 | list_del(&link->cont_link_list); | |
387 | kfree(link); | |
388 | } | |
389 | } | |
390 | ||
391 | /* | |
392 | * find_css_set() takes an existing cgroup group and a | |
393 | * cgroup object, and returns a css_set object that's | |
394 | * equivalent to the old group, but with the given cgroup | |
395 | * substituted into the appropriate hierarchy. Must be called with | |
396 | * cgroup_mutex held | |
397 | */ | |
398 | ||
399 | static struct css_set *find_css_set( | |
400 | struct css_set *oldcg, struct cgroup *cont) | |
401 | { | |
402 | struct css_set *res; | |
403 | struct cgroup_subsys_state *template[CGROUP_SUBSYS_COUNT]; | |
404 | int i; | |
405 | ||
406 | struct list_head tmp_cg_links; | |
407 | struct cg_cgroup_link *link; | |
408 | ||
409 | /* First see if we already have a cgroup group that matches | |
410 | * the desired set */ | |
411 | write_lock(&css_set_lock); | |
412 | res = find_existing_css_set(oldcg, cont, template); | |
413 | if (res) | |
414 | get_css_set(res); | |
415 | write_unlock(&css_set_lock); | |
416 | ||
417 | if (res) | |
418 | return res; | |
419 | ||
420 | res = kmalloc(sizeof(*res), GFP_KERNEL); | |
421 | if (!res) | |
422 | return NULL; | |
423 | ||
424 | /* Allocate all the cg_cgroup_link objects that we'll need */ | |
425 | if (allocate_cg_links(root_count, &tmp_cg_links) < 0) { | |
426 | kfree(res); | |
427 | return NULL; | |
428 | } | |
429 | ||
430 | kref_init(&res->ref); | |
431 | INIT_LIST_HEAD(&res->cg_links); | |
432 | INIT_LIST_HEAD(&res->tasks); | |
433 | ||
434 | /* Copy the set of subsystem state objects generated in | |
435 | * find_existing_css_set() */ | |
436 | memcpy(res->subsys, template, sizeof(res->subsys)); | |
437 | ||
438 | write_lock(&css_set_lock); | |
439 | /* Add reference counts and links from the new css_set. */ | |
440 | for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) { | |
441 | struct cgroup *cont = res->subsys[i]->cgroup; | |
442 | struct cgroup_subsys *ss = subsys[i]; | |
443 | atomic_inc(&cont->count); | |
444 | /* | |
445 | * We want to add a link once per cgroup, so we | |
446 | * only do it for the first subsystem in each | |
447 | * hierarchy | |
448 | */ | |
449 | if (ss->root->subsys_list.next == &ss->sibling) { | |
450 | BUG_ON(list_empty(&tmp_cg_links)); | |
451 | link = list_entry(tmp_cg_links.next, | |
452 | struct cg_cgroup_link, | |
453 | cont_link_list); | |
454 | list_del(&link->cont_link_list); | |
455 | list_add(&link->cont_link_list, &cont->css_sets); | |
456 | link->cg = res; | |
457 | list_add(&link->cg_link_list, &res->cg_links); | |
458 | } | |
459 | } | |
460 | if (list_empty(&rootnode.subsys_list)) { | |
461 | link = list_entry(tmp_cg_links.next, | |
462 | struct cg_cgroup_link, | |
463 | cont_link_list); | |
464 | list_del(&link->cont_link_list); | |
465 | list_add(&link->cont_link_list, &dummytop->css_sets); | |
466 | link->cg = res; | |
467 | list_add(&link->cg_link_list, &res->cg_links); | |
468 | } | |
469 | ||
470 | BUG_ON(!list_empty(&tmp_cg_links)); | |
471 | ||
472 | /* Link this cgroup group into the list */ | |
473 | list_add(&res->list, &init_css_set.list); | |
474 | css_set_count++; | |
475 | INIT_LIST_HEAD(&res->tasks); | |
476 | write_unlock(&css_set_lock); | |
477 | ||
478 | return res; | |
b4f48b63 PM |
479 | } |
480 | ||
ddbcc7e8 PM |
481 | /* |
482 | * There is one global cgroup mutex. We also require taking | |
483 | * task_lock() when dereferencing a task's cgroup subsys pointers. | |
484 | * See "The task_lock() exception", at the end of this comment. | |
485 | * | |
486 | * A task must hold cgroup_mutex to modify cgroups. | |
487 | * | |
488 | * Any task can increment and decrement the count field without lock. | |
489 | * So in general, code holding cgroup_mutex can't rely on the count | |
490 | * field not changing. However, if the count goes to zero, then only | |
491 | * attach_task() can increment it again. Because a count of zero | |
492 | * means that no tasks are currently attached, therefore there is no | |
493 | * way a task attached to that cgroup can fork (the other way to | |
494 | * increment the count). So code holding cgroup_mutex can safely | |
495 | * assume that if the count is zero, it will stay zero. Similarly, if | |
496 | * a task holds cgroup_mutex on a cgroup with zero count, it | |
497 | * knows that the cgroup won't be removed, as cgroup_rmdir() | |
498 | * needs that mutex. | |
499 | * | |
500 | * The cgroup_common_file_write handler for operations that modify | |
501 | * the cgroup hierarchy holds cgroup_mutex across the entire operation, | |
502 | * single threading all such cgroup modifications across the system. | |
503 | * | |
504 | * The fork and exit callbacks cgroup_fork() and cgroup_exit(), don't | |
505 | * (usually) take cgroup_mutex. These are the two most performance | |
506 | * critical pieces of code here. The exception occurs on cgroup_exit(), | |
507 | * when a task in a notify_on_release cgroup exits. Then cgroup_mutex | |
508 | * is taken, and if the cgroup count is zero, a usermode call made | |
509 | * to /sbin/cgroup_release_agent with the name of the cgroup (path | |
510 | * relative to the root of cgroup file system) as the argument. | |
511 | * | |
512 | * A cgroup can only be deleted if both its 'count' of using tasks | |
513 | * is zero, and its list of 'children' cgroups is empty. Since all | |
514 | * tasks in the system use _some_ cgroup, and since there is always at | |
515 | * least one task in the system (init, pid == 1), therefore, top_cgroup | |
516 | * always has either children cgroups and/or using tasks. So we don't | |
517 | * need a special hack to ensure that top_cgroup cannot be deleted. | |
518 | * | |
519 | * The task_lock() exception | |
520 | * | |
521 | * The need for this exception arises from the action of | |
522 | * attach_task(), which overwrites one tasks cgroup pointer with | |
523 | * another. It does so using cgroup_mutexe, however there are | |
524 | * several performance critical places that need to reference | |
525 | * task->cgroup without the expense of grabbing a system global | |
526 | * mutex. Therefore except as noted below, when dereferencing or, as | |
527 | * in attach_task(), modifying a task'ss cgroup pointer we use | |
528 | * task_lock(), which acts on a spinlock (task->alloc_lock) already in | |
529 | * the task_struct routinely used for such matters. | |
530 | * | |
531 | * P.S. One more locking exception. RCU is used to guard the | |
532 | * update of a tasks cgroup pointer by attach_task() | |
533 | */ | |
534 | ||
ddbcc7e8 PM |
535 | /** |
536 | * cgroup_lock - lock out any changes to cgroup structures | |
537 | * | |
538 | */ | |
539 | ||
540 | void cgroup_lock(void) | |
541 | { | |
542 | mutex_lock(&cgroup_mutex); | |
543 | } | |
544 | ||
545 | /** | |
546 | * cgroup_unlock - release lock on cgroup changes | |
547 | * | |
548 | * Undo the lock taken in a previous cgroup_lock() call. | |
549 | */ | |
550 | ||
551 | void cgroup_unlock(void) | |
552 | { | |
553 | mutex_unlock(&cgroup_mutex); | |
554 | } | |
555 | ||
556 | /* | |
557 | * A couple of forward declarations required, due to cyclic reference loop: | |
558 | * cgroup_mkdir -> cgroup_create -> cgroup_populate_dir -> | |
559 | * cgroup_add_file -> cgroup_create_file -> cgroup_dir_inode_operations | |
560 | * -> cgroup_mkdir. | |
561 | */ | |
562 | ||
563 | static int cgroup_mkdir(struct inode *dir, struct dentry *dentry, int mode); | |
564 | static int cgroup_rmdir(struct inode *unused_dir, struct dentry *dentry); | |
565 | static int cgroup_populate_dir(struct cgroup *cont); | |
566 | static struct inode_operations cgroup_dir_inode_operations; | |
a424316c PM |
567 | static struct file_operations proc_cgroupstats_operations; |
568 | ||
569 | static struct backing_dev_info cgroup_backing_dev_info = { | |
570 | .capabilities = BDI_CAP_NO_ACCT_DIRTY | BDI_CAP_NO_WRITEBACK, | |
571 | }; | |
ddbcc7e8 PM |
572 | |
573 | static struct inode *cgroup_new_inode(mode_t mode, struct super_block *sb) | |
574 | { | |
575 | struct inode *inode = new_inode(sb); | |
ddbcc7e8 PM |
576 | |
577 | if (inode) { | |
578 | inode->i_mode = mode; | |
579 | inode->i_uid = current->fsuid; | |
580 | inode->i_gid = current->fsgid; | |
581 | inode->i_blocks = 0; | |
582 | inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME; | |
583 | inode->i_mapping->backing_dev_info = &cgroup_backing_dev_info; | |
584 | } | |
585 | return inode; | |
586 | } | |
587 | ||
588 | static void cgroup_diput(struct dentry *dentry, struct inode *inode) | |
589 | { | |
590 | /* is dentry a directory ? if so, kfree() associated cgroup */ | |
591 | if (S_ISDIR(inode->i_mode)) { | |
592 | struct cgroup *cont = dentry->d_fsdata; | |
593 | BUG_ON(!(cgroup_is_removed(cont))); | |
81a6a5cd PM |
594 | /* It's possible for external users to be holding css |
595 | * reference counts on a cgroup; css_put() needs to | |
596 | * be able to access the cgroup after decrementing | |
597 | * the reference count in order to know if it needs to | |
598 | * queue the cgroup to be handled by the release | |
599 | * agent */ | |
600 | synchronize_rcu(); | |
ddbcc7e8 PM |
601 | kfree(cont); |
602 | } | |
603 | iput(inode); | |
604 | } | |
605 | ||
606 | static void remove_dir(struct dentry *d) | |
607 | { | |
608 | struct dentry *parent = dget(d->d_parent); | |
609 | ||
610 | d_delete(d); | |
611 | simple_rmdir(parent->d_inode, d); | |
612 | dput(parent); | |
613 | } | |
614 | ||
615 | static void cgroup_clear_directory(struct dentry *dentry) | |
616 | { | |
617 | struct list_head *node; | |
618 | ||
619 | BUG_ON(!mutex_is_locked(&dentry->d_inode->i_mutex)); | |
620 | spin_lock(&dcache_lock); | |
621 | node = dentry->d_subdirs.next; | |
622 | while (node != &dentry->d_subdirs) { | |
623 | struct dentry *d = list_entry(node, struct dentry, d_u.d_child); | |
624 | list_del_init(node); | |
625 | if (d->d_inode) { | |
626 | /* This should never be called on a cgroup | |
627 | * directory with child cgroups */ | |
628 | BUG_ON(d->d_inode->i_mode & S_IFDIR); | |
629 | d = dget_locked(d); | |
630 | spin_unlock(&dcache_lock); | |
631 | d_delete(d); | |
632 | simple_unlink(dentry->d_inode, d); | |
633 | dput(d); | |
634 | spin_lock(&dcache_lock); | |
635 | } | |
636 | node = dentry->d_subdirs.next; | |
637 | } | |
638 | spin_unlock(&dcache_lock); | |
639 | } | |
640 | ||
641 | /* | |
642 | * NOTE : the dentry must have been dget()'ed | |
643 | */ | |
644 | static void cgroup_d_remove_dir(struct dentry *dentry) | |
645 | { | |
646 | cgroup_clear_directory(dentry); | |
647 | ||
648 | spin_lock(&dcache_lock); | |
649 | list_del_init(&dentry->d_u.d_child); | |
650 | spin_unlock(&dcache_lock); | |
651 | remove_dir(dentry); | |
652 | } | |
653 | ||
654 | static int rebind_subsystems(struct cgroupfs_root *root, | |
655 | unsigned long final_bits) | |
656 | { | |
657 | unsigned long added_bits, removed_bits; | |
658 | struct cgroup *cont = &root->top_cgroup; | |
659 | int i; | |
660 | ||
661 | removed_bits = root->actual_subsys_bits & ~final_bits; | |
662 | added_bits = final_bits & ~root->actual_subsys_bits; | |
663 | /* Check that any added subsystems are currently free */ | |
664 | for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) { | |
665 | unsigned long long bit = 1ull << i; | |
666 | struct cgroup_subsys *ss = subsys[i]; | |
667 | if (!(bit & added_bits)) | |
668 | continue; | |
669 | if (ss->root != &rootnode) { | |
670 | /* Subsystem isn't free */ | |
671 | return -EBUSY; | |
672 | } | |
673 | } | |
674 | ||
675 | /* Currently we don't handle adding/removing subsystems when | |
676 | * any child cgroups exist. This is theoretically supportable | |
677 | * but involves complex error handling, so it's being left until | |
678 | * later */ | |
679 | if (!list_empty(&cont->children)) | |
680 | return -EBUSY; | |
681 | ||
682 | /* Process each subsystem */ | |
683 | for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) { | |
684 | struct cgroup_subsys *ss = subsys[i]; | |
685 | unsigned long bit = 1UL << i; | |
686 | if (bit & added_bits) { | |
687 | /* We're binding this subsystem to this hierarchy */ | |
688 | BUG_ON(cont->subsys[i]); | |
689 | BUG_ON(!dummytop->subsys[i]); | |
690 | BUG_ON(dummytop->subsys[i]->cgroup != dummytop); | |
691 | cont->subsys[i] = dummytop->subsys[i]; | |
692 | cont->subsys[i]->cgroup = cont; | |
693 | list_add(&ss->sibling, &root->subsys_list); | |
694 | rcu_assign_pointer(ss->root, root); | |
695 | if (ss->bind) | |
696 | ss->bind(ss, cont); | |
697 | ||
698 | } else if (bit & removed_bits) { | |
699 | /* We're removing this subsystem */ | |
700 | BUG_ON(cont->subsys[i] != dummytop->subsys[i]); | |
701 | BUG_ON(cont->subsys[i]->cgroup != cont); | |
702 | if (ss->bind) | |
703 | ss->bind(ss, dummytop); | |
704 | dummytop->subsys[i]->cgroup = dummytop; | |
705 | cont->subsys[i] = NULL; | |
706 | rcu_assign_pointer(subsys[i]->root, &rootnode); | |
707 | list_del(&ss->sibling); | |
708 | } else if (bit & final_bits) { | |
709 | /* Subsystem state should already exist */ | |
710 | BUG_ON(!cont->subsys[i]); | |
711 | } else { | |
712 | /* Subsystem state shouldn't exist */ | |
713 | BUG_ON(cont->subsys[i]); | |
714 | } | |
715 | } | |
716 | root->subsys_bits = root->actual_subsys_bits = final_bits; | |
717 | synchronize_rcu(); | |
718 | ||
719 | return 0; | |
720 | } | |
721 | ||
722 | static int cgroup_show_options(struct seq_file *seq, struct vfsmount *vfs) | |
723 | { | |
724 | struct cgroupfs_root *root = vfs->mnt_sb->s_fs_info; | |
725 | struct cgroup_subsys *ss; | |
726 | ||
727 | mutex_lock(&cgroup_mutex); | |
728 | for_each_subsys(root, ss) | |
729 | seq_printf(seq, ",%s", ss->name); | |
730 | if (test_bit(ROOT_NOPREFIX, &root->flags)) | |
731 | seq_puts(seq, ",noprefix"); | |
81a6a5cd PM |
732 | if (strlen(root->release_agent_path)) |
733 | seq_printf(seq, ",release_agent=%s", root->release_agent_path); | |
ddbcc7e8 PM |
734 | mutex_unlock(&cgroup_mutex); |
735 | return 0; | |
736 | } | |
737 | ||
738 | struct cgroup_sb_opts { | |
739 | unsigned long subsys_bits; | |
740 | unsigned long flags; | |
81a6a5cd | 741 | char *release_agent; |
ddbcc7e8 PM |
742 | }; |
743 | ||
744 | /* Convert a hierarchy specifier into a bitmask of subsystems and | |
745 | * flags. */ | |
746 | static int parse_cgroupfs_options(char *data, | |
747 | struct cgroup_sb_opts *opts) | |
748 | { | |
749 | char *token, *o = data ?: "all"; | |
750 | ||
751 | opts->subsys_bits = 0; | |
752 | opts->flags = 0; | |
81a6a5cd | 753 | opts->release_agent = NULL; |
ddbcc7e8 PM |
754 | |
755 | while ((token = strsep(&o, ",")) != NULL) { | |
756 | if (!*token) | |
757 | return -EINVAL; | |
758 | if (!strcmp(token, "all")) { | |
759 | opts->subsys_bits = (1 << CGROUP_SUBSYS_COUNT) - 1; | |
760 | } else if (!strcmp(token, "noprefix")) { | |
761 | set_bit(ROOT_NOPREFIX, &opts->flags); | |
81a6a5cd PM |
762 | } else if (!strncmp(token, "release_agent=", 14)) { |
763 | /* Specifying two release agents is forbidden */ | |
764 | if (opts->release_agent) | |
765 | return -EINVAL; | |
766 | opts->release_agent = kzalloc(PATH_MAX, GFP_KERNEL); | |
767 | if (!opts->release_agent) | |
768 | return -ENOMEM; | |
769 | strncpy(opts->release_agent, token + 14, PATH_MAX - 1); | |
770 | opts->release_agent[PATH_MAX - 1] = 0; | |
ddbcc7e8 PM |
771 | } else { |
772 | struct cgroup_subsys *ss; | |
773 | int i; | |
774 | for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) { | |
775 | ss = subsys[i]; | |
776 | if (!strcmp(token, ss->name)) { | |
777 | set_bit(i, &opts->subsys_bits); | |
778 | break; | |
779 | } | |
780 | } | |
781 | if (i == CGROUP_SUBSYS_COUNT) | |
782 | return -ENOENT; | |
783 | } | |
784 | } | |
785 | ||
786 | /* We can't have an empty hierarchy */ | |
787 | if (!opts->subsys_bits) | |
788 | return -EINVAL; | |
789 | ||
790 | return 0; | |
791 | } | |
792 | ||
793 | static int cgroup_remount(struct super_block *sb, int *flags, char *data) | |
794 | { | |
795 | int ret = 0; | |
796 | struct cgroupfs_root *root = sb->s_fs_info; | |
797 | struct cgroup *cont = &root->top_cgroup; | |
798 | struct cgroup_sb_opts opts; | |
799 | ||
800 | mutex_lock(&cont->dentry->d_inode->i_mutex); | |
801 | mutex_lock(&cgroup_mutex); | |
802 | ||
803 | /* See what subsystems are wanted */ | |
804 | ret = parse_cgroupfs_options(data, &opts); | |
805 | if (ret) | |
806 | goto out_unlock; | |
807 | ||
808 | /* Don't allow flags to change at remount */ | |
809 | if (opts.flags != root->flags) { | |
810 | ret = -EINVAL; | |
811 | goto out_unlock; | |
812 | } | |
813 | ||
814 | ret = rebind_subsystems(root, opts.subsys_bits); | |
815 | ||
816 | /* (re)populate subsystem files */ | |
817 | if (!ret) | |
818 | cgroup_populate_dir(cont); | |
819 | ||
81a6a5cd PM |
820 | if (opts.release_agent) |
821 | strcpy(root->release_agent_path, opts.release_agent); | |
ddbcc7e8 | 822 | out_unlock: |
81a6a5cd PM |
823 | if (opts.release_agent) |
824 | kfree(opts.release_agent); | |
ddbcc7e8 PM |
825 | mutex_unlock(&cgroup_mutex); |
826 | mutex_unlock(&cont->dentry->d_inode->i_mutex); | |
827 | return ret; | |
828 | } | |
829 | ||
830 | static struct super_operations cgroup_ops = { | |
831 | .statfs = simple_statfs, | |
832 | .drop_inode = generic_delete_inode, | |
833 | .show_options = cgroup_show_options, | |
834 | .remount_fs = cgroup_remount, | |
835 | }; | |
836 | ||
837 | static void init_cgroup_root(struct cgroupfs_root *root) | |
838 | { | |
839 | struct cgroup *cont = &root->top_cgroup; | |
840 | INIT_LIST_HEAD(&root->subsys_list); | |
841 | INIT_LIST_HEAD(&root->root_list); | |
842 | root->number_of_cgroups = 1; | |
843 | cont->root = root; | |
844 | cont->top_cgroup = cont; | |
845 | INIT_LIST_HEAD(&cont->sibling); | |
846 | INIT_LIST_HEAD(&cont->children); | |
817929ec | 847 | INIT_LIST_HEAD(&cont->css_sets); |
81a6a5cd | 848 | INIT_LIST_HEAD(&cont->release_list); |
ddbcc7e8 PM |
849 | } |
850 | ||
851 | static int cgroup_test_super(struct super_block *sb, void *data) | |
852 | { | |
853 | struct cgroupfs_root *new = data; | |
854 | struct cgroupfs_root *root = sb->s_fs_info; | |
855 | ||
856 | /* First check subsystems */ | |
857 | if (new->subsys_bits != root->subsys_bits) | |
858 | return 0; | |
859 | ||
860 | /* Next check flags */ | |
861 | if (new->flags != root->flags) | |
862 | return 0; | |
863 | ||
864 | return 1; | |
865 | } | |
866 | ||
867 | static int cgroup_set_super(struct super_block *sb, void *data) | |
868 | { | |
869 | int ret; | |
870 | struct cgroupfs_root *root = data; | |
871 | ||
872 | ret = set_anon_super(sb, NULL); | |
873 | if (ret) | |
874 | return ret; | |
875 | ||
876 | sb->s_fs_info = root; | |
877 | root->sb = sb; | |
878 | ||
879 | sb->s_blocksize = PAGE_CACHE_SIZE; | |
880 | sb->s_blocksize_bits = PAGE_CACHE_SHIFT; | |
881 | sb->s_magic = CGROUP_SUPER_MAGIC; | |
882 | sb->s_op = &cgroup_ops; | |
883 | ||
884 | return 0; | |
885 | } | |
886 | ||
887 | static int cgroup_get_rootdir(struct super_block *sb) | |
888 | { | |
889 | struct inode *inode = | |
890 | cgroup_new_inode(S_IFDIR | S_IRUGO | S_IXUGO | S_IWUSR, sb); | |
891 | struct dentry *dentry; | |
892 | ||
893 | if (!inode) | |
894 | return -ENOMEM; | |
895 | ||
896 | inode->i_op = &simple_dir_inode_operations; | |
897 | inode->i_fop = &simple_dir_operations; | |
898 | inode->i_op = &cgroup_dir_inode_operations; | |
899 | /* directories start off with i_nlink == 2 (for "." entry) */ | |
900 | inc_nlink(inode); | |
901 | dentry = d_alloc_root(inode); | |
902 | if (!dentry) { | |
903 | iput(inode); | |
904 | return -ENOMEM; | |
905 | } | |
906 | sb->s_root = dentry; | |
907 | return 0; | |
908 | } | |
909 | ||
910 | static int cgroup_get_sb(struct file_system_type *fs_type, | |
911 | int flags, const char *unused_dev_name, | |
912 | void *data, struct vfsmount *mnt) | |
913 | { | |
914 | struct cgroup_sb_opts opts; | |
915 | int ret = 0; | |
916 | struct super_block *sb; | |
917 | struct cgroupfs_root *root; | |
817929ec PM |
918 | struct list_head tmp_cg_links, *l; |
919 | INIT_LIST_HEAD(&tmp_cg_links); | |
ddbcc7e8 PM |
920 | |
921 | /* First find the desired set of subsystems */ | |
922 | ret = parse_cgroupfs_options(data, &opts); | |
81a6a5cd PM |
923 | if (ret) { |
924 | if (opts.release_agent) | |
925 | kfree(opts.release_agent); | |
ddbcc7e8 | 926 | return ret; |
81a6a5cd | 927 | } |
ddbcc7e8 PM |
928 | |
929 | root = kzalloc(sizeof(*root), GFP_KERNEL); | |
930 | if (!root) | |
931 | return -ENOMEM; | |
932 | ||
933 | init_cgroup_root(root); | |
934 | root->subsys_bits = opts.subsys_bits; | |
935 | root->flags = opts.flags; | |
81a6a5cd PM |
936 | if (opts.release_agent) { |
937 | strcpy(root->release_agent_path, opts.release_agent); | |
938 | kfree(opts.release_agent); | |
939 | } | |
ddbcc7e8 PM |
940 | |
941 | sb = sget(fs_type, cgroup_test_super, cgroup_set_super, root); | |
942 | ||
943 | if (IS_ERR(sb)) { | |
944 | kfree(root); | |
945 | return PTR_ERR(sb); | |
946 | } | |
947 | ||
948 | if (sb->s_fs_info != root) { | |
949 | /* Reusing an existing superblock */ | |
950 | BUG_ON(sb->s_root == NULL); | |
951 | kfree(root); | |
952 | root = NULL; | |
953 | } else { | |
954 | /* New superblock */ | |
955 | struct cgroup *cont = &root->top_cgroup; | |
817929ec | 956 | struct inode *inode; |
ddbcc7e8 PM |
957 | |
958 | BUG_ON(sb->s_root != NULL); | |
959 | ||
960 | ret = cgroup_get_rootdir(sb); | |
961 | if (ret) | |
962 | goto drop_new_super; | |
817929ec | 963 | inode = sb->s_root->d_inode; |
ddbcc7e8 | 964 | |
817929ec | 965 | mutex_lock(&inode->i_mutex); |
ddbcc7e8 PM |
966 | mutex_lock(&cgroup_mutex); |
967 | ||
817929ec PM |
968 | /* |
969 | * We're accessing css_set_count without locking | |
970 | * css_set_lock here, but that's OK - it can only be | |
971 | * increased by someone holding cgroup_lock, and | |
972 | * that's us. The worst that can happen is that we | |
973 | * have some link structures left over | |
974 | */ | |
975 | ret = allocate_cg_links(css_set_count, &tmp_cg_links); | |
976 | if (ret) { | |
977 | mutex_unlock(&cgroup_mutex); | |
978 | mutex_unlock(&inode->i_mutex); | |
979 | goto drop_new_super; | |
980 | } | |
981 | ||
ddbcc7e8 PM |
982 | ret = rebind_subsystems(root, root->subsys_bits); |
983 | if (ret == -EBUSY) { | |
984 | mutex_unlock(&cgroup_mutex); | |
817929ec | 985 | mutex_unlock(&inode->i_mutex); |
ddbcc7e8 PM |
986 | goto drop_new_super; |
987 | } | |
988 | ||
989 | /* EBUSY should be the only error here */ | |
990 | BUG_ON(ret); | |
991 | ||
992 | list_add(&root->root_list, &roots); | |
817929ec | 993 | root_count++; |
ddbcc7e8 PM |
994 | |
995 | sb->s_root->d_fsdata = &root->top_cgroup; | |
996 | root->top_cgroup.dentry = sb->s_root; | |
997 | ||
817929ec PM |
998 | /* Link the top cgroup in this hierarchy into all |
999 | * the css_set objects */ | |
1000 | write_lock(&css_set_lock); | |
1001 | l = &init_css_set.list; | |
1002 | do { | |
1003 | struct css_set *cg; | |
1004 | struct cg_cgroup_link *link; | |
1005 | cg = list_entry(l, struct css_set, list); | |
1006 | BUG_ON(list_empty(&tmp_cg_links)); | |
1007 | link = list_entry(tmp_cg_links.next, | |
1008 | struct cg_cgroup_link, | |
1009 | cont_link_list); | |
1010 | list_del(&link->cont_link_list); | |
1011 | link->cg = cg; | |
1012 | list_add(&link->cont_link_list, | |
1013 | &root->top_cgroup.css_sets); | |
1014 | list_add(&link->cg_link_list, &cg->cg_links); | |
1015 | l = l->next; | |
1016 | } while (l != &init_css_set.list); | |
1017 | write_unlock(&css_set_lock); | |
1018 | ||
1019 | free_cg_links(&tmp_cg_links); | |
1020 | ||
ddbcc7e8 PM |
1021 | BUG_ON(!list_empty(&cont->sibling)); |
1022 | BUG_ON(!list_empty(&cont->children)); | |
1023 | BUG_ON(root->number_of_cgroups != 1); | |
1024 | ||
ddbcc7e8 | 1025 | cgroup_populate_dir(cont); |
817929ec | 1026 | mutex_unlock(&inode->i_mutex); |
ddbcc7e8 PM |
1027 | mutex_unlock(&cgroup_mutex); |
1028 | } | |
1029 | ||
1030 | return simple_set_mnt(mnt, sb); | |
1031 | ||
1032 | drop_new_super: | |
1033 | up_write(&sb->s_umount); | |
1034 | deactivate_super(sb); | |
817929ec | 1035 | free_cg_links(&tmp_cg_links); |
ddbcc7e8 PM |
1036 | return ret; |
1037 | } | |
1038 | ||
1039 | static void cgroup_kill_sb(struct super_block *sb) { | |
1040 | struct cgroupfs_root *root = sb->s_fs_info; | |
1041 | struct cgroup *cont = &root->top_cgroup; | |
1042 | int ret; | |
1043 | ||
1044 | BUG_ON(!root); | |
1045 | ||
1046 | BUG_ON(root->number_of_cgroups != 1); | |
1047 | BUG_ON(!list_empty(&cont->children)); | |
1048 | BUG_ON(!list_empty(&cont->sibling)); | |
1049 | ||
1050 | mutex_lock(&cgroup_mutex); | |
1051 | ||
1052 | /* Rebind all subsystems back to the default hierarchy */ | |
1053 | ret = rebind_subsystems(root, 0); | |
1054 | /* Shouldn't be able to fail ... */ | |
1055 | BUG_ON(ret); | |
1056 | ||
817929ec PM |
1057 | /* |
1058 | * Release all the links from css_sets to this hierarchy's | |
1059 | * root cgroup | |
1060 | */ | |
1061 | write_lock(&css_set_lock); | |
1062 | while (!list_empty(&cont->css_sets)) { | |
1063 | struct cg_cgroup_link *link; | |
1064 | link = list_entry(cont->css_sets.next, | |
1065 | struct cg_cgroup_link, cont_link_list); | |
1066 | list_del(&link->cg_link_list); | |
1067 | list_del(&link->cont_link_list); | |
1068 | kfree(link); | |
1069 | } | |
1070 | write_unlock(&css_set_lock); | |
1071 | ||
1072 | if (!list_empty(&root->root_list)) { | |
ddbcc7e8 | 1073 | list_del(&root->root_list); |
817929ec PM |
1074 | root_count--; |
1075 | } | |
ddbcc7e8 PM |
1076 | mutex_unlock(&cgroup_mutex); |
1077 | ||
1078 | kfree(root); | |
1079 | kill_litter_super(sb); | |
1080 | } | |
1081 | ||
1082 | static struct file_system_type cgroup_fs_type = { | |
1083 | .name = "cgroup", | |
1084 | .get_sb = cgroup_get_sb, | |
1085 | .kill_sb = cgroup_kill_sb, | |
1086 | }; | |
1087 | ||
1088 | static inline struct cgroup *__d_cont(struct dentry *dentry) | |
1089 | { | |
1090 | return dentry->d_fsdata; | |
1091 | } | |
1092 | ||
1093 | static inline struct cftype *__d_cft(struct dentry *dentry) | |
1094 | { | |
1095 | return dentry->d_fsdata; | |
1096 | } | |
1097 | ||
1098 | /* | |
1099 | * Called with cgroup_mutex held. Writes path of cgroup into buf. | |
1100 | * Returns 0 on success, -errno on error. | |
1101 | */ | |
1102 | int cgroup_path(const struct cgroup *cont, char *buf, int buflen) | |
1103 | { | |
1104 | char *start; | |
1105 | ||
1106 | if (cont == dummytop) { | |
1107 | /* | |
1108 | * Inactive subsystems have no dentry for their root | |
1109 | * cgroup | |
1110 | */ | |
1111 | strcpy(buf, "/"); | |
1112 | return 0; | |
1113 | } | |
1114 | ||
1115 | start = buf + buflen; | |
1116 | ||
1117 | *--start = '\0'; | |
1118 | for (;;) { | |
1119 | int len = cont->dentry->d_name.len; | |
1120 | if ((start -= len) < buf) | |
1121 | return -ENAMETOOLONG; | |
1122 | memcpy(start, cont->dentry->d_name.name, len); | |
1123 | cont = cont->parent; | |
1124 | if (!cont) | |
1125 | break; | |
1126 | if (!cont->parent) | |
1127 | continue; | |
1128 | if (--start < buf) | |
1129 | return -ENAMETOOLONG; | |
1130 | *start = '/'; | |
1131 | } | |
1132 | memmove(buf, start, buf + buflen - start); | |
1133 | return 0; | |
1134 | } | |
1135 | ||
bbcb81d0 PM |
1136 | /* |
1137 | * Return the first subsystem attached to a cgroup's hierarchy, and | |
1138 | * its subsystem id. | |
1139 | */ | |
1140 | ||
1141 | static void get_first_subsys(const struct cgroup *cont, | |
1142 | struct cgroup_subsys_state **css, int *subsys_id) | |
1143 | { | |
1144 | const struct cgroupfs_root *root = cont->root; | |
1145 | const struct cgroup_subsys *test_ss; | |
1146 | BUG_ON(list_empty(&root->subsys_list)); | |
1147 | test_ss = list_entry(root->subsys_list.next, | |
1148 | struct cgroup_subsys, sibling); | |
1149 | if (css) { | |
1150 | *css = cont->subsys[test_ss->subsys_id]; | |
1151 | BUG_ON(!*css); | |
1152 | } | |
1153 | if (subsys_id) | |
1154 | *subsys_id = test_ss->subsys_id; | |
1155 | } | |
1156 | ||
1157 | /* | |
1158 | * Attach task 'tsk' to cgroup 'cont' | |
1159 | * | |
1160 | * Call holding cgroup_mutex. May take task_lock of | |
1161 | * the task 'pid' during call. | |
1162 | */ | |
1163 | static int attach_task(struct cgroup *cont, struct task_struct *tsk) | |
1164 | { | |
1165 | int retval = 0; | |
1166 | struct cgroup_subsys *ss; | |
1167 | struct cgroup *oldcont; | |
817929ec PM |
1168 | struct css_set *cg = tsk->cgroups; |
1169 | struct css_set *newcg; | |
bbcb81d0 | 1170 | struct cgroupfs_root *root = cont->root; |
bbcb81d0 PM |
1171 | int subsys_id; |
1172 | ||
1173 | get_first_subsys(cont, NULL, &subsys_id); | |
1174 | ||
1175 | /* Nothing to do if the task is already in that cgroup */ | |
1176 | oldcont = task_cgroup(tsk, subsys_id); | |
1177 | if (cont == oldcont) | |
1178 | return 0; | |
1179 | ||
1180 | for_each_subsys(root, ss) { | |
1181 | if (ss->can_attach) { | |
1182 | retval = ss->can_attach(ss, cont, tsk); | |
1183 | if (retval) { | |
1184 | return retval; | |
1185 | } | |
1186 | } | |
1187 | } | |
1188 | ||
817929ec PM |
1189 | /* |
1190 | * Locate or allocate a new css_set for this task, | |
1191 | * based on its final set of cgroups | |
1192 | */ | |
1193 | newcg = find_css_set(cg, cont); | |
1194 | if (!newcg) { | |
1195 | return -ENOMEM; | |
1196 | } | |
1197 | ||
bbcb81d0 PM |
1198 | task_lock(tsk); |
1199 | if (tsk->flags & PF_EXITING) { | |
1200 | task_unlock(tsk); | |
817929ec | 1201 | put_css_set(newcg); |
bbcb81d0 PM |
1202 | return -ESRCH; |
1203 | } | |
817929ec | 1204 | rcu_assign_pointer(tsk->cgroups, newcg); |
bbcb81d0 PM |
1205 | task_unlock(tsk); |
1206 | ||
817929ec PM |
1207 | /* Update the css_set linked lists if we're using them */ |
1208 | write_lock(&css_set_lock); | |
1209 | if (!list_empty(&tsk->cg_list)) { | |
1210 | list_del(&tsk->cg_list); | |
1211 | list_add(&tsk->cg_list, &newcg->tasks); | |
1212 | } | |
1213 | write_unlock(&css_set_lock); | |
1214 | ||
bbcb81d0 PM |
1215 | for_each_subsys(root, ss) { |
1216 | if (ss->attach) { | |
1217 | ss->attach(ss, cont, oldcont, tsk); | |
1218 | } | |
1219 | } | |
81a6a5cd | 1220 | set_bit(CONT_RELEASABLE, &oldcont->flags); |
bbcb81d0 | 1221 | synchronize_rcu(); |
817929ec | 1222 | put_css_set(cg); |
bbcb81d0 PM |
1223 | return 0; |
1224 | } | |
1225 | ||
1226 | /* | |
1227 | * Attach task with pid 'pid' to cgroup 'cont'. Call with | |
1228 | * cgroup_mutex, may take task_lock of task | |
1229 | */ | |
1230 | static int attach_task_by_pid(struct cgroup *cont, char *pidbuf) | |
1231 | { | |
1232 | pid_t pid; | |
1233 | struct task_struct *tsk; | |
1234 | int ret; | |
1235 | ||
1236 | if (sscanf(pidbuf, "%d", &pid) != 1) | |
1237 | return -EIO; | |
1238 | ||
1239 | if (pid) { | |
1240 | rcu_read_lock(); | |
1241 | tsk = find_task_by_pid(pid); | |
1242 | if (!tsk || tsk->flags & PF_EXITING) { | |
1243 | rcu_read_unlock(); | |
1244 | return -ESRCH; | |
1245 | } | |
1246 | get_task_struct(tsk); | |
1247 | rcu_read_unlock(); | |
1248 | ||
1249 | if ((current->euid) && (current->euid != tsk->uid) | |
1250 | && (current->euid != tsk->suid)) { | |
1251 | put_task_struct(tsk); | |
1252 | return -EACCES; | |
1253 | } | |
1254 | } else { | |
1255 | tsk = current; | |
1256 | get_task_struct(tsk); | |
1257 | } | |
1258 | ||
1259 | ret = attach_task(cont, tsk); | |
1260 | put_task_struct(tsk); | |
1261 | return ret; | |
1262 | } | |
1263 | ||
ddbcc7e8 PM |
1264 | /* The various types of files and directories in a cgroup file system */ |
1265 | ||
1266 | enum cgroup_filetype { | |
1267 | FILE_ROOT, | |
1268 | FILE_DIR, | |
1269 | FILE_TASKLIST, | |
81a6a5cd PM |
1270 | FILE_NOTIFY_ON_RELEASE, |
1271 | FILE_RELEASABLE, | |
1272 | FILE_RELEASE_AGENT, | |
ddbcc7e8 PM |
1273 | }; |
1274 | ||
355e0c48 PM |
1275 | static ssize_t cgroup_write_uint(struct cgroup *cont, struct cftype *cft, |
1276 | struct file *file, | |
1277 | const char __user *userbuf, | |
1278 | size_t nbytes, loff_t *unused_ppos) | |
1279 | { | |
1280 | char buffer[64]; | |
1281 | int retval = 0; | |
1282 | u64 val; | |
1283 | char *end; | |
1284 | ||
1285 | if (!nbytes) | |
1286 | return -EINVAL; | |
1287 | if (nbytes >= sizeof(buffer)) | |
1288 | return -E2BIG; | |
1289 | if (copy_from_user(buffer, userbuf, nbytes)) | |
1290 | return -EFAULT; | |
1291 | ||
1292 | buffer[nbytes] = 0; /* nul-terminate */ | |
1293 | ||
1294 | /* strip newline if necessary */ | |
1295 | if (nbytes && (buffer[nbytes-1] == '\n')) | |
1296 | buffer[nbytes-1] = 0; | |
1297 | val = simple_strtoull(buffer, &end, 0); | |
1298 | if (*end) | |
1299 | return -EINVAL; | |
1300 | ||
1301 | /* Pass to subsystem */ | |
1302 | retval = cft->write_uint(cont, cft, val); | |
1303 | if (!retval) | |
1304 | retval = nbytes; | |
1305 | return retval; | |
1306 | } | |
1307 | ||
bbcb81d0 PM |
1308 | static ssize_t cgroup_common_file_write(struct cgroup *cont, |
1309 | struct cftype *cft, | |
1310 | struct file *file, | |
1311 | const char __user *userbuf, | |
1312 | size_t nbytes, loff_t *unused_ppos) | |
1313 | { | |
1314 | enum cgroup_filetype type = cft->private; | |
1315 | char *buffer; | |
1316 | int retval = 0; | |
1317 | ||
1318 | if (nbytes >= PATH_MAX) | |
1319 | return -E2BIG; | |
1320 | ||
1321 | /* +1 for nul-terminator */ | |
1322 | buffer = kmalloc(nbytes + 1, GFP_KERNEL); | |
1323 | if (buffer == NULL) | |
1324 | return -ENOMEM; | |
1325 | ||
1326 | if (copy_from_user(buffer, userbuf, nbytes)) { | |
1327 | retval = -EFAULT; | |
1328 | goto out1; | |
1329 | } | |
1330 | buffer[nbytes] = 0; /* nul-terminate */ | |
1331 | ||
1332 | mutex_lock(&cgroup_mutex); | |
1333 | ||
1334 | if (cgroup_is_removed(cont)) { | |
1335 | retval = -ENODEV; | |
1336 | goto out2; | |
1337 | } | |
1338 | ||
1339 | switch (type) { | |
1340 | case FILE_TASKLIST: | |
1341 | retval = attach_task_by_pid(cont, buffer); | |
1342 | break; | |
81a6a5cd PM |
1343 | case FILE_NOTIFY_ON_RELEASE: |
1344 | clear_bit(CONT_RELEASABLE, &cont->flags); | |
1345 | if (simple_strtoul(buffer, NULL, 10) != 0) | |
1346 | set_bit(CONT_NOTIFY_ON_RELEASE, &cont->flags); | |
1347 | else | |
1348 | clear_bit(CONT_NOTIFY_ON_RELEASE, &cont->flags); | |
1349 | break; | |
1350 | case FILE_RELEASE_AGENT: | |
1351 | { | |
1352 | struct cgroupfs_root *root = cont->root; | |
1353 | /* Strip trailing newline */ | |
1354 | if (nbytes && (buffer[nbytes-1] == '\n')) { | |
1355 | buffer[nbytes-1] = 0; | |
1356 | } | |
1357 | if (nbytes < sizeof(root->release_agent_path)) { | |
1358 | /* We never write anything other than '\0' | |
1359 | * into the last char of release_agent_path, | |
1360 | * so it always remains a NUL-terminated | |
1361 | * string */ | |
1362 | strncpy(root->release_agent_path, buffer, nbytes); | |
1363 | root->release_agent_path[nbytes] = 0; | |
1364 | } else { | |
1365 | retval = -ENOSPC; | |
1366 | } | |
1367 | break; | |
1368 | } | |
bbcb81d0 PM |
1369 | default: |
1370 | retval = -EINVAL; | |
1371 | goto out2; | |
1372 | } | |
1373 | ||
1374 | if (retval == 0) | |
1375 | retval = nbytes; | |
1376 | out2: | |
1377 | mutex_unlock(&cgroup_mutex); | |
1378 | out1: | |
1379 | kfree(buffer); | |
1380 | return retval; | |
1381 | } | |
1382 | ||
ddbcc7e8 PM |
1383 | static ssize_t cgroup_file_write(struct file *file, const char __user *buf, |
1384 | size_t nbytes, loff_t *ppos) | |
1385 | { | |
1386 | struct cftype *cft = __d_cft(file->f_dentry); | |
1387 | struct cgroup *cont = __d_cont(file->f_dentry->d_parent); | |
1388 | ||
1389 | if (!cft) | |
1390 | return -ENODEV; | |
355e0c48 PM |
1391 | if (cft->write) |
1392 | return cft->write(cont, cft, file, buf, nbytes, ppos); | |
1393 | if (cft->write_uint) | |
1394 | return cgroup_write_uint(cont, cft, file, buf, nbytes, ppos); | |
1395 | return -EINVAL; | |
ddbcc7e8 PM |
1396 | } |
1397 | ||
1398 | static ssize_t cgroup_read_uint(struct cgroup *cont, struct cftype *cft, | |
1399 | struct file *file, | |
1400 | char __user *buf, size_t nbytes, | |
1401 | loff_t *ppos) | |
1402 | { | |
1403 | char tmp[64]; | |
1404 | u64 val = cft->read_uint(cont, cft); | |
1405 | int len = sprintf(tmp, "%llu\n", (unsigned long long) val); | |
1406 | ||
1407 | return simple_read_from_buffer(buf, nbytes, ppos, tmp, len); | |
1408 | } | |
1409 | ||
81a6a5cd PM |
1410 | static ssize_t cgroup_common_file_read(struct cgroup *cont, |
1411 | struct cftype *cft, | |
1412 | struct file *file, | |
1413 | char __user *buf, | |
1414 | size_t nbytes, loff_t *ppos) | |
1415 | { | |
1416 | enum cgroup_filetype type = cft->private; | |
1417 | char *page; | |
1418 | ssize_t retval = 0; | |
1419 | char *s; | |
1420 | ||
1421 | if (!(page = (char *)__get_free_page(GFP_KERNEL))) | |
1422 | return -ENOMEM; | |
1423 | ||
1424 | s = page; | |
1425 | ||
1426 | switch (type) { | |
1427 | case FILE_RELEASE_AGENT: | |
1428 | { | |
1429 | struct cgroupfs_root *root; | |
1430 | size_t n; | |
1431 | mutex_lock(&cgroup_mutex); | |
1432 | root = cont->root; | |
1433 | n = strnlen(root->release_agent_path, | |
1434 | sizeof(root->release_agent_path)); | |
1435 | n = min(n, (size_t) PAGE_SIZE); | |
1436 | strncpy(s, root->release_agent_path, n); | |
1437 | mutex_unlock(&cgroup_mutex); | |
1438 | s += n; | |
1439 | break; | |
1440 | } | |
1441 | default: | |
1442 | retval = -EINVAL; | |
1443 | goto out; | |
1444 | } | |
1445 | *s++ = '\n'; | |
1446 | ||
1447 | retval = simple_read_from_buffer(buf, nbytes, ppos, page, s - page); | |
1448 | out: | |
1449 | free_page((unsigned long)page); | |
1450 | return retval; | |
1451 | } | |
1452 | ||
ddbcc7e8 PM |
1453 | static ssize_t cgroup_file_read(struct file *file, char __user *buf, |
1454 | size_t nbytes, loff_t *ppos) | |
1455 | { | |
1456 | struct cftype *cft = __d_cft(file->f_dentry); | |
1457 | struct cgroup *cont = __d_cont(file->f_dentry->d_parent); | |
1458 | ||
1459 | if (!cft) | |
1460 | return -ENODEV; | |
1461 | ||
1462 | if (cft->read) | |
1463 | return cft->read(cont, cft, file, buf, nbytes, ppos); | |
1464 | if (cft->read_uint) | |
1465 | return cgroup_read_uint(cont, cft, file, buf, nbytes, ppos); | |
1466 | return -EINVAL; | |
1467 | } | |
1468 | ||
1469 | static int cgroup_file_open(struct inode *inode, struct file *file) | |
1470 | { | |
1471 | int err; | |
1472 | struct cftype *cft; | |
1473 | ||
1474 | err = generic_file_open(inode, file); | |
1475 | if (err) | |
1476 | return err; | |
1477 | ||
1478 | cft = __d_cft(file->f_dentry); | |
1479 | if (!cft) | |
1480 | return -ENODEV; | |
1481 | if (cft->open) | |
1482 | err = cft->open(inode, file); | |
1483 | else | |
1484 | err = 0; | |
1485 | ||
1486 | return err; | |
1487 | } | |
1488 | ||
1489 | static int cgroup_file_release(struct inode *inode, struct file *file) | |
1490 | { | |
1491 | struct cftype *cft = __d_cft(file->f_dentry); | |
1492 | if (cft->release) | |
1493 | return cft->release(inode, file); | |
1494 | return 0; | |
1495 | } | |
1496 | ||
1497 | /* | |
1498 | * cgroup_rename - Only allow simple rename of directories in place. | |
1499 | */ | |
1500 | static int cgroup_rename(struct inode *old_dir, struct dentry *old_dentry, | |
1501 | struct inode *new_dir, struct dentry *new_dentry) | |
1502 | { | |
1503 | if (!S_ISDIR(old_dentry->d_inode->i_mode)) | |
1504 | return -ENOTDIR; | |
1505 | if (new_dentry->d_inode) | |
1506 | return -EEXIST; | |
1507 | if (old_dir != new_dir) | |
1508 | return -EIO; | |
1509 | return simple_rename(old_dir, old_dentry, new_dir, new_dentry); | |
1510 | } | |
1511 | ||
1512 | static struct file_operations cgroup_file_operations = { | |
1513 | .read = cgroup_file_read, | |
1514 | .write = cgroup_file_write, | |
1515 | .llseek = generic_file_llseek, | |
1516 | .open = cgroup_file_open, | |
1517 | .release = cgroup_file_release, | |
1518 | }; | |
1519 | ||
1520 | static struct inode_operations cgroup_dir_inode_operations = { | |
1521 | .lookup = simple_lookup, | |
1522 | .mkdir = cgroup_mkdir, | |
1523 | .rmdir = cgroup_rmdir, | |
1524 | .rename = cgroup_rename, | |
1525 | }; | |
1526 | ||
1527 | static int cgroup_create_file(struct dentry *dentry, int mode, | |
1528 | struct super_block *sb) | |
1529 | { | |
1530 | static struct dentry_operations cgroup_dops = { | |
1531 | .d_iput = cgroup_diput, | |
1532 | }; | |
1533 | ||
1534 | struct inode *inode; | |
1535 | ||
1536 | if (!dentry) | |
1537 | return -ENOENT; | |
1538 | if (dentry->d_inode) | |
1539 | return -EEXIST; | |
1540 | ||
1541 | inode = cgroup_new_inode(mode, sb); | |
1542 | if (!inode) | |
1543 | return -ENOMEM; | |
1544 | ||
1545 | if (S_ISDIR(mode)) { | |
1546 | inode->i_op = &cgroup_dir_inode_operations; | |
1547 | inode->i_fop = &simple_dir_operations; | |
1548 | ||
1549 | /* start off with i_nlink == 2 (for "." entry) */ | |
1550 | inc_nlink(inode); | |
1551 | ||
1552 | /* start with the directory inode held, so that we can | |
1553 | * populate it without racing with another mkdir */ | |
817929ec | 1554 | mutex_lock_nested(&inode->i_mutex, I_MUTEX_CHILD); |
ddbcc7e8 PM |
1555 | } else if (S_ISREG(mode)) { |
1556 | inode->i_size = 0; | |
1557 | inode->i_fop = &cgroup_file_operations; | |
1558 | } | |
1559 | dentry->d_op = &cgroup_dops; | |
1560 | d_instantiate(dentry, inode); | |
1561 | dget(dentry); /* Extra count - pin the dentry in core */ | |
1562 | return 0; | |
1563 | } | |
1564 | ||
1565 | /* | |
1566 | * cgroup_create_dir - create a directory for an object. | |
1567 | * cont: the cgroup we create the directory for. | |
1568 | * It must have a valid ->parent field | |
1569 | * And we are going to fill its ->dentry field. | |
1570 | * dentry: dentry of the new container | |
1571 | * mode: mode to set on new directory. | |
1572 | */ | |
1573 | static int cgroup_create_dir(struct cgroup *cont, struct dentry *dentry, | |
1574 | int mode) | |
1575 | { | |
1576 | struct dentry *parent; | |
1577 | int error = 0; | |
1578 | ||
1579 | parent = cont->parent->dentry; | |
1580 | error = cgroup_create_file(dentry, S_IFDIR | mode, cont->root->sb); | |
1581 | if (!error) { | |
1582 | dentry->d_fsdata = cont; | |
1583 | inc_nlink(parent->d_inode); | |
1584 | cont->dentry = dentry; | |
1585 | dget(dentry); | |
1586 | } | |
1587 | dput(dentry); | |
1588 | ||
1589 | return error; | |
1590 | } | |
1591 | ||
1592 | int cgroup_add_file(struct cgroup *cont, | |
1593 | struct cgroup_subsys *subsys, | |
1594 | const struct cftype *cft) | |
1595 | { | |
1596 | struct dentry *dir = cont->dentry; | |
1597 | struct dentry *dentry; | |
1598 | int error; | |
1599 | ||
1600 | char name[MAX_CGROUP_TYPE_NAMELEN + MAX_CFTYPE_NAME + 2] = { 0 }; | |
1601 | if (subsys && !test_bit(ROOT_NOPREFIX, &cont->root->flags)) { | |
1602 | strcpy(name, subsys->name); | |
1603 | strcat(name, "."); | |
1604 | } | |
1605 | strcat(name, cft->name); | |
1606 | BUG_ON(!mutex_is_locked(&dir->d_inode->i_mutex)); | |
1607 | dentry = lookup_one_len(name, dir, strlen(name)); | |
1608 | if (!IS_ERR(dentry)) { | |
1609 | error = cgroup_create_file(dentry, 0644 | S_IFREG, | |
1610 | cont->root->sb); | |
1611 | if (!error) | |
1612 | dentry->d_fsdata = (void *)cft; | |
1613 | dput(dentry); | |
1614 | } else | |
1615 | error = PTR_ERR(dentry); | |
1616 | return error; | |
1617 | } | |
1618 | ||
1619 | int cgroup_add_files(struct cgroup *cont, | |
1620 | struct cgroup_subsys *subsys, | |
1621 | const struct cftype cft[], | |
1622 | int count) | |
1623 | { | |
1624 | int i, err; | |
1625 | for (i = 0; i < count; i++) { | |
1626 | err = cgroup_add_file(cont, subsys, &cft[i]); | |
1627 | if (err) | |
1628 | return err; | |
1629 | } | |
1630 | return 0; | |
1631 | } | |
1632 | ||
817929ec PM |
1633 | /* Count the number of tasks in a cgroup. */ |
1634 | ||
1635 | int cgroup_task_count(const struct cgroup *cont) | |
bbcb81d0 PM |
1636 | { |
1637 | int count = 0; | |
817929ec PM |
1638 | struct list_head *l; |
1639 | ||
1640 | read_lock(&css_set_lock); | |
1641 | l = cont->css_sets.next; | |
1642 | while (l != &cont->css_sets) { | |
1643 | struct cg_cgroup_link *link = | |
1644 | list_entry(l, struct cg_cgroup_link, cont_link_list); | |
1645 | count += atomic_read(&link->cg->ref.refcount); | |
1646 | l = l->next; | |
1647 | } | |
1648 | read_unlock(&css_set_lock); | |
bbcb81d0 PM |
1649 | return count; |
1650 | } | |
1651 | ||
817929ec PM |
1652 | /* |
1653 | * Advance a list_head iterator. The iterator should be positioned at | |
1654 | * the start of a css_set | |
1655 | */ | |
1656 | static void cgroup_advance_iter(struct cgroup *cont, | |
1657 | struct cgroup_iter *it) | |
1658 | { | |
1659 | struct list_head *l = it->cg_link; | |
1660 | struct cg_cgroup_link *link; | |
1661 | struct css_set *cg; | |
1662 | ||
1663 | /* Advance to the next non-empty css_set */ | |
1664 | do { | |
1665 | l = l->next; | |
1666 | if (l == &cont->css_sets) { | |
1667 | it->cg_link = NULL; | |
1668 | return; | |
1669 | } | |
1670 | link = list_entry(l, struct cg_cgroup_link, cont_link_list); | |
1671 | cg = link->cg; | |
1672 | } while (list_empty(&cg->tasks)); | |
1673 | it->cg_link = l; | |
1674 | it->task = cg->tasks.next; | |
1675 | } | |
1676 | ||
1677 | void cgroup_iter_start(struct cgroup *cont, struct cgroup_iter *it) | |
1678 | { | |
1679 | /* | |
1680 | * The first time anyone tries to iterate across a cgroup, | |
1681 | * we need to enable the list linking each css_set to its | |
1682 | * tasks, and fix up all existing tasks. | |
1683 | */ | |
1684 | if (!use_task_css_set_links) { | |
1685 | struct task_struct *p, *g; | |
1686 | write_lock(&css_set_lock); | |
1687 | use_task_css_set_links = 1; | |
1688 | do_each_thread(g, p) { | |
1689 | task_lock(p); | |
1690 | if (list_empty(&p->cg_list)) | |
1691 | list_add(&p->cg_list, &p->cgroups->tasks); | |
1692 | task_unlock(p); | |
1693 | } while_each_thread(g, p); | |
1694 | write_unlock(&css_set_lock); | |
1695 | } | |
1696 | read_lock(&css_set_lock); | |
1697 | it->cg_link = &cont->css_sets; | |
1698 | cgroup_advance_iter(cont, it); | |
1699 | } | |
1700 | ||
1701 | struct task_struct *cgroup_iter_next(struct cgroup *cont, | |
1702 | struct cgroup_iter *it) | |
1703 | { | |
1704 | struct task_struct *res; | |
1705 | struct list_head *l = it->task; | |
1706 | ||
1707 | /* If the iterator cg is NULL, we have no tasks */ | |
1708 | if (!it->cg_link) | |
1709 | return NULL; | |
1710 | res = list_entry(l, struct task_struct, cg_list); | |
1711 | /* Advance iterator to find next entry */ | |
1712 | l = l->next; | |
1713 | if (l == &res->cgroups->tasks) { | |
1714 | /* We reached the end of this task list - move on to | |
1715 | * the next cg_cgroup_link */ | |
1716 | cgroup_advance_iter(cont, it); | |
1717 | } else { | |
1718 | it->task = l; | |
1719 | } | |
1720 | return res; | |
1721 | } | |
1722 | ||
1723 | void cgroup_iter_end(struct cgroup *cont, struct cgroup_iter *it) | |
1724 | { | |
1725 | read_unlock(&css_set_lock); | |
1726 | } | |
1727 | ||
bbcb81d0 PM |
1728 | /* |
1729 | * Stuff for reading the 'tasks' file. | |
1730 | * | |
1731 | * Reading this file can return large amounts of data if a cgroup has | |
1732 | * *lots* of attached tasks. So it may need several calls to read(), | |
1733 | * but we cannot guarantee that the information we produce is correct | |
1734 | * unless we produce it entirely atomically. | |
1735 | * | |
1736 | * Upon tasks file open(), a struct ctr_struct is allocated, that | |
1737 | * will have a pointer to an array (also allocated here). The struct | |
1738 | * ctr_struct * is stored in file->private_data. Its resources will | |
1739 | * be freed by release() when the file is closed. The array is used | |
1740 | * to sprintf the PIDs and then used by read(). | |
1741 | */ | |
1742 | struct ctr_struct { | |
1743 | char *buf; | |
1744 | int bufsz; | |
1745 | }; | |
1746 | ||
1747 | /* | |
1748 | * Load into 'pidarray' up to 'npids' of the tasks using cgroup | |
1749 | * 'cont'. Return actual number of pids loaded. No need to | |
1750 | * task_lock(p) when reading out p->cgroup, since we're in an RCU | |
1751 | * read section, so the css_set can't go away, and is | |
1752 | * immutable after creation. | |
1753 | */ | |
1754 | static int pid_array_load(pid_t *pidarray, int npids, struct cgroup *cont) | |
1755 | { | |
1756 | int n = 0; | |
817929ec PM |
1757 | struct cgroup_iter it; |
1758 | struct task_struct *tsk; | |
1759 | cgroup_iter_start(cont, &it); | |
1760 | while ((tsk = cgroup_iter_next(cont, &it))) { | |
1761 | if (unlikely(n == npids)) | |
1762 | break; | |
1763 | pidarray[n++] = pid_nr(task_pid(tsk)); | |
1764 | } | |
1765 | cgroup_iter_end(cont, &it); | |
bbcb81d0 PM |
1766 | return n; |
1767 | } | |
1768 | ||
1769 | static int cmppid(const void *a, const void *b) | |
1770 | { | |
1771 | return *(pid_t *)a - *(pid_t *)b; | |
1772 | } | |
1773 | ||
1774 | /* | |
1775 | * Convert array 'a' of 'npids' pid_t's to a string of newline separated | |
1776 | * decimal pids in 'buf'. Don't write more than 'sz' chars, but return | |
1777 | * count 'cnt' of how many chars would be written if buf were large enough. | |
1778 | */ | |
1779 | static int pid_array_to_buf(char *buf, int sz, pid_t *a, int npids) | |
1780 | { | |
1781 | int cnt = 0; | |
1782 | int i; | |
1783 | ||
1784 | for (i = 0; i < npids; i++) | |
1785 | cnt += snprintf(buf + cnt, max(sz - cnt, 0), "%d\n", a[i]); | |
1786 | return cnt; | |
1787 | } | |
1788 | ||
1789 | /* | |
1790 | * Handle an open on 'tasks' file. Prepare a buffer listing the | |
1791 | * process id's of tasks currently attached to the cgroup being opened. | |
1792 | * | |
1793 | * Does not require any specific cgroup mutexes, and does not take any. | |
1794 | */ | |
1795 | static int cgroup_tasks_open(struct inode *unused, struct file *file) | |
1796 | { | |
1797 | struct cgroup *cont = __d_cont(file->f_dentry->d_parent); | |
1798 | struct ctr_struct *ctr; | |
1799 | pid_t *pidarray; | |
1800 | int npids; | |
1801 | char c; | |
1802 | ||
1803 | if (!(file->f_mode & FMODE_READ)) | |
1804 | return 0; | |
1805 | ||
1806 | ctr = kmalloc(sizeof(*ctr), GFP_KERNEL); | |
1807 | if (!ctr) | |
1808 | goto err0; | |
1809 | ||
1810 | /* | |
1811 | * If cgroup gets more users after we read count, we won't have | |
1812 | * enough space - tough. This race is indistinguishable to the | |
1813 | * caller from the case that the additional cgroup users didn't | |
1814 | * show up until sometime later on. | |
1815 | */ | |
1816 | npids = cgroup_task_count(cont); | |
1817 | if (npids) { | |
1818 | pidarray = kmalloc(npids * sizeof(pid_t), GFP_KERNEL); | |
1819 | if (!pidarray) | |
1820 | goto err1; | |
1821 | ||
1822 | npids = pid_array_load(pidarray, npids, cont); | |
1823 | sort(pidarray, npids, sizeof(pid_t), cmppid, NULL); | |
1824 | ||
1825 | /* Call pid_array_to_buf() twice, first just to get bufsz */ | |
1826 | ctr->bufsz = pid_array_to_buf(&c, sizeof(c), pidarray, npids) + 1; | |
1827 | ctr->buf = kmalloc(ctr->bufsz, GFP_KERNEL); | |
1828 | if (!ctr->buf) | |
1829 | goto err2; | |
1830 | ctr->bufsz = pid_array_to_buf(ctr->buf, ctr->bufsz, pidarray, npids); | |
1831 | ||
1832 | kfree(pidarray); | |
1833 | } else { | |
1834 | ctr->buf = 0; | |
1835 | ctr->bufsz = 0; | |
1836 | } | |
1837 | file->private_data = ctr; | |
1838 | return 0; | |
1839 | ||
1840 | err2: | |
1841 | kfree(pidarray); | |
1842 | err1: | |
1843 | kfree(ctr); | |
1844 | err0: | |
1845 | return -ENOMEM; | |
1846 | } | |
1847 | ||
1848 | static ssize_t cgroup_tasks_read(struct cgroup *cont, | |
1849 | struct cftype *cft, | |
1850 | struct file *file, char __user *buf, | |
1851 | size_t nbytes, loff_t *ppos) | |
1852 | { | |
1853 | struct ctr_struct *ctr = file->private_data; | |
1854 | ||
1855 | return simple_read_from_buffer(buf, nbytes, ppos, ctr->buf, ctr->bufsz); | |
1856 | } | |
1857 | ||
1858 | static int cgroup_tasks_release(struct inode *unused_inode, | |
1859 | struct file *file) | |
1860 | { | |
1861 | struct ctr_struct *ctr; | |
1862 | ||
1863 | if (file->f_mode & FMODE_READ) { | |
1864 | ctr = file->private_data; | |
1865 | kfree(ctr->buf); | |
1866 | kfree(ctr); | |
1867 | } | |
1868 | return 0; | |
1869 | } | |
1870 | ||
81a6a5cd PM |
1871 | static u64 cgroup_read_notify_on_release(struct cgroup *cont, |
1872 | struct cftype *cft) | |
1873 | { | |
1874 | return notify_on_release(cont); | |
1875 | } | |
1876 | ||
1877 | static u64 cgroup_read_releasable(struct cgroup *cont, struct cftype *cft) | |
1878 | { | |
1879 | return test_bit(CONT_RELEASABLE, &cont->flags); | |
1880 | } | |
1881 | ||
bbcb81d0 PM |
1882 | /* |
1883 | * for the common functions, 'private' gives the type of file | |
1884 | */ | |
81a6a5cd PM |
1885 | static struct cftype files[] = { |
1886 | { | |
1887 | .name = "tasks", | |
1888 | .open = cgroup_tasks_open, | |
1889 | .read = cgroup_tasks_read, | |
1890 | .write = cgroup_common_file_write, | |
1891 | .release = cgroup_tasks_release, | |
1892 | .private = FILE_TASKLIST, | |
1893 | }, | |
1894 | ||
1895 | { | |
1896 | .name = "notify_on_release", | |
1897 | .read_uint = cgroup_read_notify_on_release, | |
1898 | .write = cgroup_common_file_write, | |
1899 | .private = FILE_NOTIFY_ON_RELEASE, | |
1900 | }, | |
1901 | ||
1902 | { | |
1903 | .name = "releasable", | |
1904 | .read_uint = cgroup_read_releasable, | |
1905 | .private = FILE_RELEASABLE, | |
1906 | } | |
1907 | }; | |
1908 | ||
1909 | static struct cftype cft_release_agent = { | |
1910 | .name = "release_agent", | |
1911 | .read = cgroup_common_file_read, | |
bbcb81d0 | 1912 | .write = cgroup_common_file_write, |
81a6a5cd | 1913 | .private = FILE_RELEASE_AGENT, |
bbcb81d0 PM |
1914 | }; |
1915 | ||
ddbcc7e8 PM |
1916 | static int cgroup_populate_dir(struct cgroup *cont) |
1917 | { | |
1918 | int err; | |
1919 | struct cgroup_subsys *ss; | |
1920 | ||
1921 | /* First clear out any existing files */ | |
1922 | cgroup_clear_directory(cont->dentry); | |
1923 | ||
81a6a5cd | 1924 | err = cgroup_add_files(cont, NULL, files, ARRAY_SIZE(files)); |
bbcb81d0 PM |
1925 | if (err < 0) |
1926 | return err; | |
1927 | ||
81a6a5cd PM |
1928 | if (cont == cont->top_cgroup) { |
1929 | if ((err = cgroup_add_file(cont, NULL, &cft_release_agent)) < 0) | |
1930 | return err; | |
1931 | } | |
1932 | ||
ddbcc7e8 PM |
1933 | for_each_subsys(cont->root, ss) { |
1934 | if (ss->populate && (err = ss->populate(ss, cont)) < 0) | |
1935 | return err; | |
1936 | } | |
1937 | ||
1938 | return 0; | |
1939 | } | |
1940 | ||
1941 | static void init_cgroup_css(struct cgroup_subsys_state *css, | |
1942 | struct cgroup_subsys *ss, | |
1943 | struct cgroup *cont) | |
1944 | { | |
1945 | css->cgroup = cont; | |
1946 | atomic_set(&css->refcnt, 0); | |
1947 | css->flags = 0; | |
1948 | if (cont == dummytop) | |
1949 | set_bit(CSS_ROOT, &css->flags); | |
1950 | BUG_ON(cont->subsys[ss->subsys_id]); | |
1951 | cont->subsys[ss->subsys_id] = css; | |
1952 | } | |
1953 | ||
1954 | /* | |
1955 | * cgroup_create - create a cgroup | |
1956 | * parent: cgroup that will be parent of the new cgroup. | |
1957 | * name: name of the new cgroup. Will be strcpy'ed. | |
1958 | * mode: mode to set on new inode | |
1959 | * | |
1960 | * Must be called with the mutex on the parent inode held | |
1961 | */ | |
1962 | ||
1963 | static long cgroup_create(struct cgroup *parent, struct dentry *dentry, | |
1964 | int mode) | |
1965 | { | |
1966 | struct cgroup *cont; | |
1967 | struct cgroupfs_root *root = parent->root; | |
1968 | int err = 0; | |
1969 | struct cgroup_subsys *ss; | |
1970 | struct super_block *sb = root->sb; | |
1971 | ||
1972 | cont = kzalloc(sizeof(*cont), GFP_KERNEL); | |
1973 | if (!cont) | |
1974 | return -ENOMEM; | |
1975 | ||
1976 | /* Grab a reference on the superblock so the hierarchy doesn't | |
1977 | * get deleted on unmount if there are child cgroups. This | |
1978 | * can be done outside cgroup_mutex, since the sb can't | |
1979 | * disappear while someone has an open control file on the | |
1980 | * fs */ | |
1981 | atomic_inc(&sb->s_active); | |
1982 | ||
1983 | mutex_lock(&cgroup_mutex); | |
1984 | ||
1985 | cont->flags = 0; | |
1986 | INIT_LIST_HEAD(&cont->sibling); | |
1987 | INIT_LIST_HEAD(&cont->children); | |
817929ec | 1988 | INIT_LIST_HEAD(&cont->css_sets); |
81a6a5cd | 1989 | INIT_LIST_HEAD(&cont->release_list); |
ddbcc7e8 PM |
1990 | |
1991 | cont->parent = parent; | |
1992 | cont->root = parent->root; | |
1993 | cont->top_cgroup = parent->top_cgroup; | |
1994 | ||
1995 | for_each_subsys(root, ss) { | |
1996 | struct cgroup_subsys_state *css = ss->create(ss, cont); | |
1997 | if (IS_ERR(css)) { | |
1998 | err = PTR_ERR(css); | |
1999 | goto err_destroy; | |
2000 | } | |
2001 | init_cgroup_css(css, ss, cont); | |
2002 | } | |
2003 | ||
2004 | list_add(&cont->sibling, &cont->parent->children); | |
2005 | root->number_of_cgroups++; | |
2006 | ||
2007 | err = cgroup_create_dir(cont, dentry, mode); | |
2008 | if (err < 0) | |
2009 | goto err_remove; | |
2010 | ||
2011 | /* The cgroup directory was pre-locked for us */ | |
2012 | BUG_ON(!mutex_is_locked(&cont->dentry->d_inode->i_mutex)); | |
2013 | ||
2014 | err = cgroup_populate_dir(cont); | |
2015 | /* If err < 0, we have a half-filled directory - oh well ;) */ | |
2016 | ||
2017 | mutex_unlock(&cgroup_mutex); | |
2018 | mutex_unlock(&cont->dentry->d_inode->i_mutex); | |
2019 | ||
2020 | return 0; | |
2021 | ||
2022 | err_remove: | |
2023 | ||
2024 | list_del(&cont->sibling); | |
2025 | root->number_of_cgroups--; | |
2026 | ||
2027 | err_destroy: | |
2028 | ||
2029 | for_each_subsys(root, ss) { | |
2030 | if (cont->subsys[ss->subsys_id]) | |
2031 | ss->destroy(ss, cont); | |
2032 | } | |
2033 | ||
2034 | mutex_unlock(&cgroup_mutex); | |
2035 | ||
2036 | /* Release the reference count that we took on the superblock */ | |
2037 | deactivate_super(sb); | |
2038 | ||
2039 | kfree(cont); | |
2040 | return err; | |
2041 | } | |
2042 | ||
2043 | static int cgroup_mkdir(struct inode *dir, struct dentry *dentry, int mode) | |
2044 | { | |
2045 | struct cgroup *c_parent = dentry->d_parent->d_fsdata; | |
2046 | ||
2047 | /* the vfs holds inode->i_mutex already */ | |
2048 | return cgroup_create(c_parent, dentry, mode | S_IFDIR); | |
2049 | } | |
2050 | ||
81a6a5cd PM |
2051 | static inline int cgroup_has_css_refs(struct cgroup *cont) |
2052 | { | |
2053 | /* Check the reference count on each subsystem. Since we | |
2054 | * already established that there are no tasks in the | |
2055 | * cgroup, if the css refcount is also 0, then there should | |
2056 | * be no outstanding references, so the subsystem is safe to | |
2057 | * destroy. We scan across all subsystems rather than using | |
2058 | * the per-hierarchy linked list of mounted subsystems since | |
2059 | * we can be called via check_for_release() with no | |
2060 | * synchronization other than RCU, and the subsystem linked | |
2061 | * list isn't RCU-safe */ | |
2062 | int i; | |
2063 | for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) { | |
2064 | struct cgroup_subsys *ss = subsys[i]; | |
2065 | struct cgroup_subsys_state *css; | |
2066 | /* Skip subsystems not in this hierarchy */ | |
2067 | if (ss->root != cont->root) | |
2068 | continue; | |
2069 | css = cont->subsys[ss->subsys_id]; | |
2070 | /* When called from check_for_release() it's possible | |
2071 | * that by this point the cgroup has been removed | |
2072 | * and the css deleted. But a false-positive doesn't | |
2073 | * matter, since it can only happen if the cgroup | |
2074 | * has been deleted and hence no longer needs the | |
2075 | * release agent to be called anyway. */ | |
2076 | if (css && atomic_read(&css->refcnt)) { | |
2077 | return 1; | |
2078 | } | |
2079 | } | |
2080 | return 0; | |
2081 | } | |
2082 | ||
ddbcc7e8 PM |
2083 | static int cgroup_rmdir(struct inode *unused_dir, struct dentry *dentry) |
2084 | { | |
2085 | struct cgroup *cont = dentry->d_fsdata; | |
2086 | struct dentry *d; | |
2087 | struct cgroup *parent; | |
2088 | struct cgroup_subsys *ss; | |
2089 | struct super_block *sb; | |
2090 | struct cgroupfs_root *root; | |
ddbcc7e8 PM |
2091 | |
2092 | /* the vfs holds both inode->i_mutex already */ | |
2093 | ||
2094 | mutex_lock(&cgroup_mutex); | |
2095 | if (atomic_read(&cont->count) != 0) { | |
2096 | mutex_unlock(&cgroup_mutex); | |
2097 | return -EBUSY; | |
2098 | } | |
2099 | if (!list_empty(&cont->children)) { | |
2100 | mutex_unlock(&cgroup_mutex); | |
2101 | return -EBUSY; | |
2102 | } | |
2103 | ||
2104 | parent = cont->parent; | |
2105 | root = cont->root; | |
2106 | sb = root->sb; | |
2107 | ||
81a6a5cd | 2108 | if (cgroup_has_css_refs(cont)) { |
ddbcc7e8 PM |
2109 | mutex_unlock(&cgroup_mutex); |
2110 | return -EBUSY; | |
2111 | } | |
2112 | ||
2113 | for_each_subsys(root, ss) { | |
2114 | if (cont->subsys[ss->subsys_id]) | |
2115 | ss->destroy(ss, cont); | |
2116 | } | |
2117 | ||
81a6a5cd | 2118 | spin_lock(&release_list_lock); |
ddbcc7e8 | 2119 | set_bit(CONT_REMOVED, &cont->flags); |
81a6a5cd PM |
2120 | if (!list_empty(&cont->release_list)) |
2121 | list_del(&cont->release_list); | |
2122 | spin_unlock(&release_list_lock); | |
ddbcc7e8 PM |
2123 | /* delete my sibling from parent->children */ |
2124 | list_del(&cont->sibling); | |
2125 | spin_lock(&cont->dentry->d_lock); | |
2126 | d = dget(cont->dentry); | |
2127 | cont->dentry = NULL; | |
2128 | spin_unlock(&d->d_lock); | |
2129 | ||
2130 | cgroup_d_remove_dir(d); | |
2131 | dput(d); | |
2132 | root->number_of_cgroups--; | |
2133 | ||
81a6a5cd PM |
2134 | set_bit(CONT_RELEASABLE, &parent->flags); |
2135 | check_for_release(parent); | |
2136 | ||
ddbcc7e8 PM |
2137 | mutex_unlock(&cgroup_mutex); |
2138 | /* Drop the active superblock reference that we took when we | |
2139 | * created the cgroup */ | |
2140 | deactivate_super(sb); | |
2141 | return 0; | |
2142 | } | |
2143 | ||
2144 | static void cgroup_init_subsys(struct cgroup_subsys *ss) | |
2145 | { | |
ddbcc7e8 | 2146 | struct cgroup_subsys_state *css; |
817929ec | 2147 | struct list_head *l; |
ddbcc7e8 PM |
2148 | printk(KERN_ERR "Initializing cgroup subsys %s\n", ss->name); |
2149 | ||
2150 | /* Create the top cgroup state for this subsystem */ | |
2151 | ss->root = &rootnode; | |
2152 | css = ss->create(ss, dummytop); | |
2153 | /* We don't handle early failures gracefully */ | |
2154 | BUG_ON(IS_ERR(css)); | |
2155 | init_cgroup_css(css, ss, dummytop); | |
2156 | ||
817929ec PM |
2157 | /* Update all cgroup groups to contain a subsys |
2158 | * pointer to this state - since the subsystem is | |
2159 | * newly registered, all tasks and hence all cgroup | |
2160 | * groups are in the subsystem's top cgroup. */ | |
2161 | write_lock(&css_set_lock); | |
2162 | l = &init_css_set.list; | |
2163 | do { | |
2164 | struct css_set *cg = | |
2165 | list_entry(l, struct css_set, list); | |
2166 | cg->subsys[ss->subsys_id] = dummytop->subsys[ss->subsys_id]; | |
2167 | l = l->next; | |
2168 | } while (l != &init_css_set.list); | |
2169 | write_unlock(&css_set_lock); | |
ddbcc7e8 PM |
2170 | |
2171 | /* If this subsystem requested that it be notified with fork | |
2172 | * events, we should send it one now for every process in the | |
2173 | * system */ | |
81a6a5cd PM |
2174 | if (ss->fork) { |
2175 | struct task_struct *g, *p; | |
2176 | ||
2177 | read_lock(&tasklist_lock); | |
2178 | do_each_thread(g, p) { | |
2179 | ss->fork(ss, p); | |
2180 | } while_each_thread(g, p); | |
2181 | read_unlock(&tasklist_lock); | |
2182 | } | |
ddbcc7e8 PM |
2183 | |
2184 | need_forkexit_callback |= ss->fork || ss->exit; | |
2185 | ||
2186 | ss->active = 1; | |
2187 | } | |
2188 | ||
2189 | /** | |
2190 | * cgroup_init_early - initialize cgroups at system boot, and | |
2191 | * initialize any subsystems that request early init. | |
2192 | */ | |
2193 | int __init cgroup_init_early(void) | |
2194 | { | |
2195 | int i; | |
817929ec PM |
2196 | kref_init(&init_css_set.ref); |
2197 | kref_get(&init_css_set.ref); | |
2198 | INIT_LIST_HEAD(&init_css_set.list); | |
2199 | INIT_LIST_HEAD(&init_css_set.cg_links); | |
2200 | INIT_LIST_HEAD(&init_css_set.tasks); | |
2201 | css_set_count = 1; | |
ddbcc7e8 PM |
2202 | init_cgroup_root(&rootnode); |
2203 | list_add(&rootnode.root_list, &roots); | |
817929ec PM |
2204 | root_count = 1; |
2205 | init_task.cgroups = &init_css_set; | |
2206 | ||
2207 | init_css_set_link.cg = &init_css_set; | |
2208 | list_add(&init_css_set_link.cont_link_list, | |
2209 | &rootnode.top_cgroup.css_sets); | |
2210 | list_add(&init_css_set_link.cg_link_list, | |
2211 | &init_css_set.cg_links); | |
ddbcc7e8 PM |
2212 | |
2213 | for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) { | |
2214 | struct cgroup_subsys *ss = subsys[i]; | |
2215 | ||
2216 | BUG_ON(!ss->name); | |
2217 | BUG_ON(strlen(ss->name) > MAX_CGROUP_TYPE_NAMELEN); | |
2218 | BUG_ON(!ss->create); | |
2219 | BUG_ON(!ss->destroy); | |
2220 | if (ss->subsys_id != i) { | |
2221 | printk(KERN_ERR "Subsys %s id == %d\n", | |
2222 | ss->name, ss->subsys_id); | |
2223 | BUG(); | |
2224 | } | |
2225 | ||
2226 | if (ss->early_init) | |
2227 | cgroup_init_subsys(ss); | |
2228 | } | |
2229 | return 0; | |
2230 | } | |
2231 | ||
2232 | /** | |
2233 | * cgroup_init - register cgroup filesystem and /proc file, and | |
2234 | * initialize any subsystems that didn't request early init. | |
2235 | */ | |
2236 | int __init cgroup_init(void) | |
2237 | { | |
2238 | int err; | |
2239 | int i; | |
a424316c PM |
2240 | struct proc_dir_entry *entry; |
2241 | ||
2242 | err = bdi_init(&cgroup_backing_dev_info); | |
2243 | if (err) | |
2244 | return err; | |
ddbcc7e8 PM |
2245 | |
2246 | for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) { | |
2247 | struct cgroup_subsys *ss = subsys[i]; | |
2248 | if (!ss->early_init) | |
2249 | cgroup_init_subsys(ss); | |
2250 | } | |
2251 | ||
2252 | err = register_filesystem(&cgroup_fs_type); | |
2253 | if (err < 0) | |
2254 | goto out; | |
2255 | ||
a424316c PM |
2256 | entry = create_proc_entry("cgroups", 0, NULL); |
2257 | if (entry) | |
2258 | entry->proc_fops = &proc_cgroupstats_operations; | |
2259 | ||
ddbcc7e8 | 2260 | out: |
a424316c PM |
2261 | if (err) |
2262 | bdi_destroy(&cgroup_backing_dev_info); | |
2263 | ||
ddbcc7e8 PM |
2264 | return err; |
2265 | } | |
b4f48b63 | 2266 | |
a424316c PM |
2267 | /* |
2268 | * proc_cgroup_show() | |
2269 | * - Print task's cgroup paths into seq_file, one line for each hierarchy | |
2270 | * - Used for /proc/<pid>/cgroup. | |
2271 | * - No need to task_lock(tsk) on this tsk->cgroup reference, as it | |
2272 | * doesn't really matter if tsk->cgroup changes after we read it, | |
2273 | * and we take cgroup_mutex, keeping attach_task() from changing it | |
2274 | * anyway. No need to check that tsk->cgroup != NULL, thanks to | |
2275 | * the_top_cgroup_hack in cgroup_exit(), which sets an exiting tasks | |
2276 | * cgroup to top_cgroup. | |
2277 | */ | |
2278 | ||
2279 | /* TODO: Use a proper seq_file iterator */ | |
2280 | static int proc_cgroup_show(struct seq_file *m, void *v) | |
2281 | { | |
2282 | struct pid *pid; | |
2283 | struct task_struct *tsk; | |
2284 | char *buf; | |
2285 | int retval; | |
2286 | struct cgroupfs_root *root; | |
2287 | ||
2288 | retval = -ENOMEM; | |
2289 | buf = kmalloc(PAGE_SIZE, GFP_KERNEL); | |
2290 | if (!buf) | |
2291 | goto out; | |
2292 | ||
2293 | retval = -ESRCH; | |
2294 | pid = m->private; | |
2295 | tsk = get_pid_task(pid, PIDTYPE_PID); | |
2296 | if (!tsk) | |
2297 | goto out_free; | |
2298 | ||
2299 | retval = 0; | |
2300 | ||
2301 | mutex_lock(&cgroup_mutex); | |
2302 | ||
2303 | for_each_root(root) { | |
2304 | struct cgroup_subsys *ss; | |
2305 | struct cgroup *cont; | |
2306 | int subsys_id; | |
2307 | int count = 0; | |
2308 | ||
2309 | /* Skip this hierarchy if it has no active subsystems */ | |
2310 | if (!root->actual_subsys_bits) | |
2311 | continue; | |
2312 | for_each_subsys(root, ss) | |
2313 | seq_printf(m, "%s%s", count++ ? "," : "", ss->name); | |
2314 | seq_putc(m, ':'); | |
2315 | get_first_subsys(&root->top_cgroup, NULL, &subsys_id); | |
2316 | cont = task_cgroup(tsk, subsys_id); | |
2317 | retval = cgroup_path(cont, buf, PAGE_SIZE); | |
2318 | if (retval < 0) | |
2319 | goto out_unlock; | |
2320 | seq_puts(m, buf); | |
2321 | seq_putc(m, '\n'); | |
2322 | } | |
2323 | ||
2324 | out_unlock: | |
2325 | mutex_unlock(&cgroup_mutex); | |
2326 | put_task_struct(tsk); | |
2327 | out_free: | |
2328 | kfree(buf); | |
2329 | out: | |
2330 | return retval; | |
2331 | } | |
2332 | ||
2333 | static int cgroup_open(struct inode *inode, struct file *file) | |
2334 | { | |
2335 | struct pid *pid = PROC_I(inode)->pid; | |
2336 | return single_open(file, proc_cgroup_show, pid); | |
2337 | } | |
2338 | ||
2339 | struct file_operations proc_cgroup_operations = { | |
2340 | .open = cgroup_open, | |
2341 | .read = seq_read, | |
2342 | .llseek = seq_lseek, | |
2343 | .release = single_release, | |
2344 | }; | |
2345 | ||
2346 | /* Display information about each subsystem and each hierarchy */ | |
2347 | static int proc_cgroupstats_show(struct seq_file *m, void *v) | |
2348 | { | |
2349 | int i; | |
2350 | struct cgroupfs_root *root; | |
2351 | ||
817929ec | 2352 | seq_puts(m, "#subsys_name\thierarchy\tnum_cgroups\n"); |
a424316c | 2353 | mutex_lock(&cgroup_mutex); |
a424316c PM |
2354 | for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) { |
2355 | struct cgroup_subsys *ss = subsys[i]; | |
817929ec PM |
2356 | seq_printf(m, "%s\t%lu\t%d\n", |
2357 | ss->name, ss->root->subsys_bits, | |
2358 | ss->root->number_of_cgroups); | |
a424316c PM |
2359 | } |
2360 | mutex_unlock(&cgroup_mutex); | |
2361 | return 0; | |
2362 | } | |
2363 | ||
2364 | static int cgroupstats_open(struct inode *inode, struct file *file) | |
2365 | { | |
2366 | return single_open(file, proc_cgroupstats_show, 0); | |
2367 | } | |
2368 | ||
2369 | static struct file_operations proc_cgroupstats_operations = { | |
2370 | .open = cgroupstats_open, | |
2371 | .read = seq_read, | |
2372 | .llseek = seq_lseek, | |
2373 | .release = single_release, | |
2374 | }; | |
2375 | ||
b4f48b63 PM |
2376 | /** |
2377 | * cgroup_fork - attach newly forked task to its parents cgroup. | |
2378 | * @tsk: pointer to task_struct of forking parent process. | |
2379 | * | |
2380 | * Description: A task inherits its parent's cgroup at fork(). | |
2381 | * | |
2382 | * A pointer to the shared css_set was automatically copied in | |
2383 | * fork.c by dup_task_struct(). However, we ignore that copy, since | |
2384 | * it was not made under the protection of RCU or cgroup_mutex, so | |
2385 | * might no longer be a valid cgroup pointer. attach_task() might | |
817929ec PM |
2386 | * have already changed current->cgroups, allowing the previously |
2387 | * referenced cgroup group to be removed and freed. | |
b4f48b63 PM |
2388 | * |
2389 | * At the point that cgroup_fork() is called, 'current' is the parent | |
2390 | * task, and the passed argument 'child' points to the child task. | |
2391 | */ | |
2392 | void cgroup_fork(struct task_struct *child) | |
2393 | { | |
817929ec PM |
2394 | task_lock(current); |
2395 | child->cgroups = current->cgroups; | |
2396 | get_css_set(child->cgroups); | |
2397 | task_unlock(current); | |
2398 | INIT_LIST_HEAD(&child->cg_list); | |
b4f48b63 PM |
2399 | } |
2400 | ||
2401 | /** | |
2402 | * cgroup_fork_callbacks - called on a new task very soon before | |
2403 | * adding it to the tasklist. No need to take any locks since no-one | |
2404 | * can be operating on this task | |
2405 | */ | |
2406 | void cgroup_fork_callbacks(struct task_struct *child) | |
2407 | { | |
2408 | if (need_forkexit_callback) { | |
2409 | int i; | |
2410 | for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) { | |
2411 | struct cgroup_subsys *ss = subsys[i]; | |
2412 | if (ss->fork) | |
2413 | ss->fork(ss, child); | |
2414 | } | |
2415 | } | |
2416 | } | |
2417 | ||
817929ec PM |
2418 | /** |
2419 | * cgroup_post_fork - called on a new task after adding it to the | |
2420 | * task list. Adds the task to the list running through its css_set | |
2421 | * if necessary. Has to be after the task is visible on the task list | |
2422 | * in case we race with the first call to cgroup_iter_start() - to | |
2423 | * guarantee that the new task ends up on its list. */ | |
2424 | void cgroup_post_fork(struct task_struct *child) | |
2425 | { | |
2426 | if (use_task_css_set_links) { | |
2427 | write_lock(&css_set_lock); | |
2428 | if (list_empty(&child->cg_list)) | |
2429 | list_add(&child->cg_list, &child->cgroups->tasks); | |
2430 | write_unlock(&css_set_lock); | |
2431 | } | |
2432 | } | |
b4f48b63 PM |
2433 | /** |
2434 | * cgroup_exit - detach cgroup from exiting task | |
2435 | * @tsk: pointer to task_struct of exiting process | |
2436 | * | |
2437 | * Description: Detach cgroup from @tsk and release it. | |
2438 | * | |
2439 | * Note that cgroups marked notify_on_release force every task in | |
2440 | * them to take the global cgroup_mutex mutex when exiting. | |
2441 | * This could impact scaling on very large systems. Be reluctant to | |
2442 | * use notify_on_release cgroups where very high task exit scaling | |
2443 | * is required on large systems. | |
2444 | * | |
2445 | * the_top_cgroup_hack: | |
2446 | * | |
2447 | * Set the exiting tasks cgroup to the root cgroup (top_cgroup). | |
2448 | * | |
2449 | * We call cgroup_exit() while the task is still competent to | |
2450 | * handle notify_on_release(), then leave the task attached to the | |
2451 | * root cgroup in each hierarchy for the remainder of its exit. | |
2452 | * | |
2453 | * To do this properly, we would increment the reference count on | |
2454 | * top_cgroup, and near the very end of the kernel/exit.c do_exit() | |
2455 | * code we would add a second cgroup function call, to drop that | |
2456 | * reference. This would just create an unnecessary hot spot on | |
2457 | * the top_cgroup reference count, to no avail. | |
2458 | * | |
2459 | * Normally, holding a reference to a cgroup without bumping its | |
2460 | * count is unsafe. The cgroup could go away, or someone could | |
2461 | * attach us to a different cgroup, decrementing the count on | |
2462 | * the first cgroup that we never incremented. But in this case, | |
2463 | * top_cgroup isn't going away, and either task has PF_EXITING set, | |
2464 | * which wards off any attach_task() attempts, or task is a failed | |
2465 | * fork, never visible to attach_task. | |
2466 | * | |
2467 | */ | |
2468 | void cgroup_exit(struct task_struct *tsk, int run_callbacks) | |
2469 | { | |
2470 | int i; | |
817929ec | 2471 | struct css_set *cg; |
b4f48b63 PM |
2472 | |
2473 | if (run_callbacks && need_forkexit_callback) { | |
2474 | for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) { | |
2475 | struct cgroup_subsys *ss = subsys[i]; | |
2476 | if (ss->exit) | |
2477 | ss->exit(ss, tsk); | |
2478 | } | |
2479 | } | |
817929ec PM |
2480 | |
2481 | /* | |
2482 | * Unlink from the css_set task list if necessary. | |
2483 | * Optimistically check cg_list before taking | |
2484 | * css_set_lock | |
2485 | */ | |
2486 | if (!list_empty(&tsk->cg_list)) { | |
2487 | write_lock(&css_set_lock); | |
2488 | if (!list_empty(&tsk->cg_list)) | |
2489 | list_del(&tsk->cg_list); | |
2490 | write_unlock(&css_set_lock); | |
2491 | } | |
2492 | ||
b4f48b63 PM |
2493 | /* Reassign the task to the init_css_set. */ |
2494 | task_lock(tsk); | |
817929ec PM |
2495 | cg = tsk->cgroups; |
2496 | tsk->cgroups = &init_css_set; | |
b4f48b63 | 2497 | task_unlock(tsk); |
817929ec | 2498 | if (cg) |
81a6a5cd | 2499 | put_css_set_taskexit(cg); |
b4f48b63 | 2500 | } |
697f4161 PM |
2501 | |
2502 | /** | |
2503 | * cgroup_clone - duplicate the current cgroup in the hierarchy | |
2504 | * that the given subsystem is attached to, and move this task into | |
2505 | * the new child | |
2506 | */ | |
2507 | int cgroup_clone(struct task_struct *tsk, struct cgroup_subsys *subsys) | |
2508 | { | |
2509 | struct dentry *dentry; | |
2510 | int ret = 0; | |
2511 | char nodename[MAX_CGROUP_TYPE_NAMELEN]; | |
2512 | struct cgroup *parent, *child; | |
2513 | struct inode *inode; | |
2514 | struct css_set *cg; | |
2515 | struct cgroupfs_root *root; | |
2516 | struct cgroup_subsys *ss; | |
2517 | ||
2518 | /* We shouldn't be called by an unregistered subsystem */ | |
2519 | BUG_ON(!subsys->active); | |
2520 | ||
2521 | /* First figure out what hierarchy and cgroup we're dealing | |
2522 | * with, and pin them so we can drop cgroup_mutex */ | |
2523 | mutex_lock(&cgroup_mutex); | |
2524 | again: | |
2525 | root = subsys->root; | |
2526 | if (root == &rootnode) { | |
2527 | printk(KERN_INFO | |
2528 | "Not cloning cgroup for unused subsystem %s\n", | |
2529 | subsys->name); | |
2530 | mutex_unlock(&cgroup_mutex); | |
2531 | return 0; | |
2532 | } | |
817929ec | 2533 | cg = tsk->cgroups; |
697f4161 PM |
2534 | parent = task_cgroup(tsk, subsys->subsys_id); |
2535 | ||
2536 | snprintf(nodename, MAX_CGROUP_TYPE_NAMELEN, "node_%d", tsk->pid); | |
2537 | ||
2538 | /* Pin the hierarchy */ | |
2539 | atomic_inc(&parent->root->sb->s_active); | |
2540 | ||
817929ec PM |
2541 | /* Keep the cgroup alive */ |
2542 | get_css_set(cg); | |
697f4161 PM |
2543 | mutex_unlock(&cgroup_mutex); |
2544 | ||
2545 | /* Now do the VFS work to create a cgroup */ | |
2546 | inode = parent->dentry->d_inode; | |
2547 | ||
2548 | /* Hold the parent directory mutex across this operation to | |
2549 | * stop anyone else deleting the new cgroup */ | |
2550 | mutex_lock(&inode->i_mutex); | |
2551 | dentry = lookup_one_len(nodename, parent->dentry, strlen(nodename)); | |
2552 | if (IS_ERR(dentry)) { | |
2553 | printk(KERN_INFO | |
2554 | "Couldn't allocate dentry for %s: %ld\n", nodename, | |
2555 | PTR_ERR(dentry)); | |
2556 | ret = PTR_ERR(dentry); | |
2557 | goto out_release; | |
2558 | } | |
2559 | ||
2560 | /* Create the cgroup directory, which also creates the cgroup */ | |
2561 | ret = vfs_mkdir(inode, dentry, S_IFDIR | 0755); | |
2562 | child = __d_cont(dentry); | |
2563 | dput(dentry); | |
2564 | if (ret) { | |
2565 | printk(KERN_INFO | |
2566 | "Failed to create cgroup %s: %d\n", nodename, | |
2567 | ret); | |
2568 | goto out_release; | |
2569 | } | |
2570 | ||
2571 | if (!child) { | |
2572 | printk(KERN_INFO | |
2573 | "Couldn't find new cgroup %s\n", nodename); | |
2574 | ret = -ENOMEM; | |
2575 | goto out_release; | |
2576 | } | |
2577 | ||
2578 | /* The cgroup now exists. Retake cgroup_mutex and check | |
2579 | * that we're still in the same state that we thought we | |
2580 | * were. */ | |
2581 | mutex_lock(&cgroup_mutex); | |
2582 | if ((root != subsys->root) || | |
2583 | (parent != task_cgroup(tsk, subsys->subsys_id))) { | |
2584 | /* Aargh, we raced ... */ | |
2585 | mutex_unlock(&inode->i_mutex); | |
817929ec | 2586 | put_css_set(cg); |
697f4161 PM |
2587 | |
2588 | deactivate_super(parent->root->sb); | |
2589 | /* The cgroup is still accessible in the VFS, but | |
2590 | * we're not going to try to rmdir() it at this | |
2591 | * point. */ | |
2592 | printk(KERN_INFO | |
2593 | "Race in cgroup_clone() - leaking cgroup %s\n", | |
2594 | nodename); | |
2595 | goto again; | |
2596 | } | |
2597 | ||
2598 | /* do any required auto-setup */ | |
2599 | for_each_subsys(root, ss) { | |
2600 | if (ss->post_clone) | |
2601 | ss->post_clone(ss, child); | |
2602 | } | |
2603 | ||
2604 | /* All seems fine. Finish by moving the task into the new cgroup */ | |
2605 | ret = attach_task(child, tsk); | |
2606 | mutex_unlock(&cgroup_mutex); | |
2607 | ||
2608 | out_release: | |
2609 | mutex_unlock(&inode->i_mutex); | |
81a6a5cd PM |
2610 | |
2611 | mutex_lock(&cgroup_mutex); | |
817929ec | 2612 | put_css_set(cg); |
81a6a5cd | 2613 | mutex_unlock(&cgroup_mutex); |
697f4161 PM |
2614 | deactivate_super(parent->root->sb); |
2615 | return ret; | |
2616 | } | |
2617 | ||
2618 | /* | |
2619 | * See if "cont" is a descendant of the current task's cgroup in | |
2620 | * the appropriate hierarchy | |
2621 | * | |
2622 | * If we are sending in dummytop, then presumably we are creating | |
2623 | * the top cgroup in the subsystem. | |
2624 | * | |
2625 | * Called only by the ns (nsproxy) cgroup. | |
2626 | */ | |
2627 | int cgroup_is_descendant(const struct cgroup *cont) | |
2628 | { | |
2629 | int ret; | |
2630 | struct cgroup *target; | |
2631 | int subsys_id; | |
2632 | ||
2633 | if (cont == dummytop) | |
2634 | return 1; | |
2635 | ||
2636 | get_first_subsys(cont, NULL, &subsys_id); | |
2637 | target = task_cgroup(current, subsys_id); | |
2638 | while (cont != target && cont!= cont->top_cgroup) | |
2639 | cont = cont->parent; | |
2640 | ret = (cont == target); | |
2641 | return ret; | |
2642 | } | |
81a6a5cd PM |
2643 | |
2644 | static void check_for_release(struct cgroup *cont) | |
2645 | { | |
2646 | /* All of these checks rely on RCU to keep the cgroup | |
2647 | * structure alive */ | |
2648 | if (cgroup_is_releasable(cont) && !atomic_read(&cont->count) | |
2649 | && list_empty(&cont->children) && !cgroup_has_css_refs(cont)) { | |
2650 | /* Control Group is currently removeable. If it's not | |
2651 | * already queued for a userspace notification, queue | |
2652 | * it now */ | |
2653 | int need_schedule_work = 0; | |
2654 | spin_lock(&release_list_lock); | |
2655 | if (!cgroup_is_removed(cont) && | |
2656 | list_empty(&cont->release_list)) { | |
2657 | list_add(&cont->release_list, &release_list); | |
2658 | need_schedule_work = 1; | |
2659 | } | |
2660 | spin_unlock(&release_list_lock); | |
2661 | if (need_schedule_work) | |
2662 | schedule_work(&release_agent_work); | |
2663 | } | |
2664 | } | |
2665 | ||
2666 | void __css_put(struct cgroup_subsys_state *css) | |
2667 | { | |
2668 | struct cgroup *cont = css->cgroup; | |
2669 | rcu_read_lock(); | |
2670 | if (atomic_dec_and_test(&css->refcnt) && notify_on_release(cont)) { | |
2671 | set_bit(CONT_RELEASABLE, &cont->flags); | |
2672 | check_for_release(cont); | |
2673 | } | |
2674 | rcu_read_unlock(); | |
2675 | } | |
2676 | ||
2677 | /* | |
2678 | * Notify userspace when a cgroup is released, by running the | |
2679 | * configured release agent with the name of the cgroup (path | |
2680 | * relative to the root of cgroup file system) as the argument. | |
2681 | * | |
2682 | * Most likely, this user command will try to rmdir this cgroup. | |
2683 | * | |
2684 | * This races with the possibility that some other task will be | |
2685 | * attached to this cgroup before it is removed, or that some other | |
2686 | * user task will 'mkdir' a child cgroup of this cgroup. That's ok. | |
2687 | * The presumed 'rmdir' will fail quietly if this cgroup is no longer | |
2688 | * unused, and this cgroup will be reprieved from its death sentence, | |
2689 | * to continue to serve a useful existence. Next time it's released, | |
2690 | * we will get notified again, if it still has 'notify_on_release' set. | |
2691 | * | |
2692 | * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which | |
2693 | * means only wait until the task is successfully execve()'d. The | |
2694 | * separate release agent task is forked by call_usermodehelper(), | |
2695 | * then control in this thread returns here, without waiting for the | |
2696 | * release agent task. We don't bother to wait because the caller of | |
2697 | * this routine has no use for the exit status of the release agent | |
2698 | * task, so no sense holding our caller up for that. | |
2699 | * | |
2700 | */ | |
2701 | ||
2702 | static void cgroup_release_agent(struct work_struct *work) | |
2703 | { | |
2704 | BUG_ON(work != &release_agent_work); | |
2705 | mutex_lock(&cgroup_mutex); | |
2706 | spin_lock(&release_list_lock); | |
2707 | while (!list_empty(&release_list)) { | |
2708 | char *argv[3], *envp[3]; | |
2709 | int i; | |
2710 | char *pathbuf; | |
2711 | struct cgroup *cont = list_entry(release_list.next, | |
2712 | struct cgroup, | |
2713 | release_list); | |
2714 | list_del_init(&cont->release_list); | |
2715 | spin_unlock(&release_list_lock); | |
2716 | pathbuf = kmalloc(PAGE_SIZE, GFP_KERNEL); | |
2717 | if (!pathbuf) { | |
2718 | spin_lock(&release_list_lock); | |
2719 | continue; | |
2720 | } | |
2721 | ||
2722 | if (cgroup_path(cont, pathbuf, PAGE_SIZE) < 0) { | |
2723 | kfree(pathbuf); | |
2724 | spin_lock(&release_list_lock); | |
2725 | continue; | |
2726 | } | |
2727 | ||
2728 | i = 0; | |
2729 | argv[i++] = cont->root->release_agent_path; | |
2730 | argv[i++] = (char *)pathbuf; | |
2731 | argv[i] = NULL; | |
2732 | ||
2733 | i = 0; | |
2734 | /* minimal command environment */ | |
2735 | envp[i++] = "HOME=/"; | |
2736 | envp[i++] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin"; | |
2737 | envp[i] = NULL; | |
2738 | ||
2739 | /* Drop the lock while we invoke the usermode helper, | |
2740 | * since the exec could involve hitting disk and hence | |
2741 | * be a slow process */ | |
2742 | mutex_unlock(&cgroup_mutex); | |
2743 | call_usermodehelper(argv[0], argv, envp, UMH_WAIT_EXEC); | |
2744 | kfree(pathbuf); | |
2745 | mutex_lock(&cgroup_mutex); | |
2746 | spin_lock(&release_list_lock); | |
2747 | } | |
2748 | spin_unlock(&release_list_lock); | |
2749 | mutex_unlock(&cgroup_mutex); | |
2750 | } |